KitKat Journal · Weekly completo
Journal Semanal — 30/03/2026 → 05/04/2026
Esta edição reúne os posts completos aprovados na janela editorial. A ordem segue Gmail editorial, Substack e Skool, preservando estrutura, mídia útil e origem canônica sem resumo.
Use o índice para saltar entre os textos. Cada post mantém o corpo integral e termina com a origem e o link da fonte.
UV for Portable Python in Agent Skills
UV for Portable Python in Agent Skills - by Eleanor Berger # Elite AI Assisted Coding Elite AI Assisted Coding The ONE place to go for everything you need to know about AI assisted coding and tools Substack's , and and . ?Eleanor Berger Dec 14, 2025 5 1 Share Python scripts are a natural choice for agent skills and agentic workflows, but traditional package management creates portability problems. Scripts need to run reliably across different execution environments — your local machine with a coding agent, a remote agent environment, or someone else’s setup entirely when sharing a skill. Traditionally, Python package and environment management is a bit messy. It depends on a Python interpreter that could be installed in many different paths, plus creating a virtual environment — which is itself rather hacky, operating by creating directories and setting shell variables. These inconsistencies cause failures when skills move between environments. UV — the modern Python package manager from Astral — solves this elegantly. **A UV-based script declares its dependencies inline and runs identically everywhere, with no environment setup required.** ### How It Works UV provides two features that matter here: 1. **Interpreter management** — UV can run and, if necessary, install the correct Python interpreter, so you always work with the correct version. 2. **On-the-fly package installation** — UV fetches, installs, and caches packages automatically when running a script. When a script includes PEP 723 inline metadata, UV handles everything. The metadata lives at the top of your script: ``` # /// script= “>=3.11”= [ # “requests==2.32.5”, # “markdown==3.10”, # ] # /// import requests import markdown``` Running `uv run script.py` installs any missing packages, caches them for future runs, and executes the script — no virtual environment creation, no permanent installation, no shell manipulation. ### Applying This to Skills In your skill’s `SKILL.md`, specify that scripts should run via UV: ``` Run the foo script to produce bar: ```bash uv run scripts/foo.py ``` ``` For agent-wide consistency, add instructions to your agent’s configuration: ``` ## Python - Unless instructed otherwise, always use the `uv` Python environment and package manager for Python. - `uv run ...` for running a python script. - `uvx ...` for running a program directly from a PyPI package. - `uv pip ...` for managing environments, installing packages, etc. ``` ### Why This Matters The full dependency specification travels with the script itself. We can package a script without requiring the execution environment to do any particular installation or virtual environment setup. This means consistent behavior and consistent performance across different execution environments — exactly what portable skills require. A new user can check out your skill and run it immediately. UV handles the rest. This follows the same principle behind explicit dependency declaration in any robust system: **never rely on implicit existence of packages in the surrounding environment.** The full and explicit dependency specification should apply uniformly across all environments where the code runs. If you do need to install packages into an environment, UV handles that too — it can use its own package management format or work with `requirements.txt` and other standard formats. ### Summary By sticking to UV in skills and agentic workflows for running and managing Python scripts, you get portable, repeatable, reproducible behavior. For Python-based skills meant to be shared and reused, this portability is not optional. UV makes it practical. * * * #### to Elite AI Assisted Coding By Eleanor Berger The ONE place to go for everything you need to know about AI assisted coding and tools Substack's , and and . 5 Likes∙ [1 Restack](https://substack.com/note/p-181575302/ ?utm_source=substack&utm_content=facepile-) 5 1 Share #### Comments Top Latest Complex Agentic Coding with Copilot: GPT-5 vs Claude 4 Sonnet [OpenAI released GPT-5 yesterday, promoting it as their best model yet for agentic coding.](http
Prompts
# /// script # requires-python = “>=3.11” # dependencies = [ # “requests==2.32.5”, # “markdown==3.10”, # ] # /// import requests import markdown # ... rest of script
O Claude Code virou meu secretário executivo
O Claude Code virou meu secretário executivo # Newsletter do Moa Newsletter do Moa O meu objetivo a Newsletter do Moa é ajudar profissionais de tecnologia que desejam desenvolver uma visão mais estratégica, mostrando os bastidores da jornada de construção de um SaaS. Over 3,000 subscribers Substack's , and and . ?ário executivo ### Conheça meu plano de compartilhar TODA minha vida com a IA Moacir Moda Mar 07, 2026 8 1 Share 👋🏼 Bem vindo a edição de número #140 da _**Newsletter do Moa**_**.** * * * Eu precisava me preparar para uma entrevista importante. O candidato tinha mandado 30 minutos de áudios no WhatsApp. Pensei: “Sério que vou ouvir tudo isso de novo?”. Foi aí que baixei os arquivos, abri o Claude Code, e em minutos tinha tudo transcrito e organizado. Naquele momento, percebi que minha forma de trabalhar nunca mais seria a mesma. Na edição de hoje da Newsletter do Moa, eu conto como estou tentando transformar minha vida toda em um contexto estruturado para que a IA me ajude a tomar boas decisões. ✉️ Se você trabalha com tecnologia e deseja pensar de forma mais estratégica, junte-se a 3.832 pessoas e assine a _Newsletter do Moa!_ * * *_Claude Code_ Conversando com um dos programadores do Tintim, mencionei que estávamos buscando gente boa para o time de _marketing_. Ele mencionou que tinha um amigo que poderia ser uma boa contratação para o time. Não dei muita bola, afinal, as chances de um programador saber identificar um bom _marketeiro_ são baixas, né? Mas, para não correr o risco de comer bola, eu falei para ele passar meu contato ao seu amigo e pedir para me chamar no WhatsApp. Meu plano era simples: faço uma pré-entrevista via WhatsApp, de forma assíncrona, e, se eu notar algum potencial, chamo para um papo mais formal. E não é que a indicação foi boa? Conversando pelo WhatsApp, vi que a pessoa tinha um bom perfil, um bom _track record,_ e um bom conhecimento sobre os assuntos em questão. Então, marquei uma entrevista, de fato, para a semana seguinte. No dia, para me preparar para a entrevista, eu precisava relembrar o contexto para montar as perguntas. Pensei: “Sério que vou ter que ouvir novamente 30 minutos de áudios?”. Como um bom preguiçoso que sou, baixei todos os áudios e pedi para o Claude Code transcrever. Em cima dessas transcrições, comecei a elaborar as perguntas. Esse exemplo mostra como eu já não consigo viver sem inteligência artificial. Meu _setup_ de trabalho quase sempre possui uma janela aberta com um _chat_. Durante muito tempo, esse _chat_ era com o GPT, mas, depois de ter uma experiência completa com o Claude Code, decidi migrar meu _setup_ de trabalho para lá. Muito do meu processo de trabalho era, literalmente, copiar e colar contextos no _chat_ para conseguir uma conversa com mais qualidade e personalização. Esses contextos eram transcrições de reuniões, documentos, _sites_, etc. Era trabalhoso, mas, até então, eu não tinha notado o custo desse trabalho, pois não imaginava alternativas. Mas, depois de passar um final de semana inteiro _vibecodando_ um projetinho no Claude Code, um mundo de possibilidades se abriu na minha cabeça. A capacidade da IA de, proativamente, abrir arquivos, fazer buscas, fazer perguntas, entre outros, eleva absurdamente a experiência de uso da IA para trabalho. Desde então, eu passei a usar o Claude Code como uma das minhas principais ferramentas de trabalho. O fluxo de trabalho é simples: baixo todos os arquivos que preciso e os coloco numa pasta. Então, abro o Claude Code nessa pasta, explico que o contexto está todo ali, e passo a trabalhar já contando com esse contexto. Minha produtividade (que já tinha aumentado com IA) aumentou absurdamente depois de adotar esse fluxo. Todo o processo seletivo que fiz com esse candidato foi dessa forma. Fazia uma entrevista, transcrevia a entrevista, colocava o arquivo como contexto, e ia “pensar em voz alta” junto com o Claude. Repeti esse processo com outras atividades de conhecimento (que são praticamente as únicas atividades que trabalho hoje em dia).Como CEO de uma empresa que vem mais do que dobrando de tamanho ano a ano, meu dia a dia é repleto de desafios que eu nunca tinha enfrentado antes. Eu, que sempre fui uma pessoa acostumada a arregaçar as mangas e meter a mão na massa, ainda tenho muita dificuldade de aceitar que meu trabalho não é mais esse. Além disso, ter uma empresa de _software_ como serviço na era da IA é viver, constantemente, sob um ruído intenso. Se eu abro o WhatsApp vou encontrar, pelo menos, um grupo por dia discutindo sobre a morte do SaaS. Se eu abro o X vou encontrar, pelo menos, um _tweet_ por dia mostrando como uma nova empresa do Vale do Silício está conquistando $1B de ARR com apenas dois funcionários e um time de agentes. Some a isso os problemas recorrentes de uma operação que cresce desenfreadamente. É servidor que cai, é custo por _lead_ que sobe, é vendedor que falta porque a oitava pessoa da família morreu neste ano… Todo esse barulho, interno e externo, só contribui para aumentar a confusão na minha cabeça. Meu maior desafio, atualmente, é vigiar para não desperdiçar meu tempo. Frequentemente, me pego fazendo algo menos importante, justamente por essa falta de clareza.ão, depois de um amigo comentar que voltou a estudar programação usando IA, eu decidi fazer o mesmo: usar IA para me ajudar a buscar essa clareza. Comecei a conversar com o Claude, e ele foi me fazendo uma série de perguntas. A cada nova rodada, novas perguntas, e a cada nova pergunta, eu tendo que ligar o microfone e falar pra caramba, ou então buscar documentos e colar no _chat_. Um trabalho moroso para um preguiçoso como eu. Foi quando me veio o _insight_: > _“E se eu tivesse toda minha vida documentada num único local, e sempre trabalhasse com a IA em cima dessa documentação?”._ Eu já tenho o hábito de documentar tudo, mas de forma descentralizada e desorganizada. Se eu organizar essa informação e disponibilizar para a IA, eu terei a coisa mais inteligente do mundo conhecendo minha vida de cabo a rabo, e tendo o contexto necessário para me ajudar a tomar as melhores decisões. Pesquisando sobre o assunto, me deparei com esse vídeo, do excelente Nate B Jones, explicando como construir um “segundo cérebro” com IA. Com mais um pouco de pesquisa, cheguei ao vídeo de um cara chamado Brad, que implementou um segundo cérebro usando as _skills_ do Claude Code. Melhor que isso, ele disponibilizou o repositório_git_do projeto dele. No domingo, eu fiz o _setup_ do projeto na minha máquina e comecei a usar. Ainda não é 100% adequado para o meu fluxo de trabalho, mas já foi um excelente ponto de partida. A ideia que estou agora é simples: controlar meus projetos e tarefas, pessoais e profissionais, usando inteligência artificial. O controle desses _backlogs_ será feito através do processamento dos contextos que já existem no meu dia a dia de trabalho: _reports_ em vídeo, reuniões de _follow-up_, reuniões e documentos de trabalho, entre outros. Ao processar esses documentos, o Claude vai, sozinho, atualizando o estado das coisas. Com todo o estado atual da minha vida documentado e atualizado, a inteligência artificial se torna uma ferramenta poderosíssima, capaz de me ajudar a tomar decisões. A ideia é usar a IA para agir como uma espécie de _chief of staff_, ou secretária executiva. Já adianto que a ideia é bem embrionária e que eu não tenho a mínima ideia se isso vai dar certo ou não. Em todo caso, decidi compartilhar com você essa minha aspiração, por dois motivos: 1) quero te inspirar a repensar seus processos de trabalho, partindo desse novo paradigma que a era da inteligência artificial trouxe, e 2) quero saber se você está fazendo algo similar e deseja compartilhar seus aprendizados comigo. Comente aqui abaixo suas opiniões e sua experiência. Duas cabeças pensam melhor que uma (três, se levarmos em conta a IA). Share * * * > `🫡 Quer minha ajuda para alavancar seu negócio? Veja como posso te ajudar:` #### **🗣️ Quer divulgar seu negócio para mais de 3.832 empreendedores, programadores e profissionais de tecnologia?** Então patrocine a Newsletter do Moa! Toque aqui para saber mais detalhes. #### **🕵🏻♂️ Está com dificuldade de identificar as vendas feitas pelo WhatsApp?** Use o Tintim para identificar e rastrear as vendas feitas pelo WhatsApp. Com o Tintim você rastreia suas conversas e acessa insights valiosos para otimizar suas campanhas. Toque aqui para fazer um teste grátis. #### **💪🏻 Quer trabalhar diretamente comigo no 1-1?** Eu resolvi abrir algumas poucas vagas para mentorias individuais. Se você quer meu acompanhamento para desenvolver seu negócio e se desenvolver como empreendedor de tecnologia, toque aqui e faça sua aplicação. * * * ##### 🤩 Quer trabalhar diretamente comigo no Tintim? Então toque aqui e confira nosso painel de vagas! . ##### 👀 Me siga nas redes sociais: Instagram | Youtube | LinkedIn | Twitter . ##### **🎥 Vídeo da semana no Canal do Moa:** . ##### **Como foi a minha semana?** 🏋🏻♀️ Pratiquei 5 dias de exercício físico e completei 49 dos 250 dias da meta do ano. 📚 Estudei 4 dias e completei 35 dos 200 dias da meta do ano. 📈 Toque aqui e veja a minha planilha de acompanhamento de métricas. . ##### **O que eu estou lendo:** 📚 Estou lendo 7 Rules of Power. 📚 Estou lendo McMafia: A Journey Through the Global Criminal Underworld. 🗂️ Toque aqui para saber quais os livros já li e a nota que dou a eles. . ##### **O que eu consumi essa semana que gostei e recomendo?** 📺 O excelente papo com o Lucas Abreu no Talks by Leo. . * * * 🫡 Esta foi mais uma edição da _**Newsletter do Moa**_**.** 👋🏼 Bem vindos aos novos leitores desde a semana passada. Já somos **3.832**profissionais estratégicos! POLL ### O que achou desta edição? 🤩 😐 😨 👊🏼 O meu objetivo com esta _newsletter_ é ajudar profissionais de tecnologia que desejam desenvolver uma visão mais estratégica. Além disso, pretendo
😅 March: Lost
😅 March: Lost - by Naomi from Todoist - Todoist Newsletterseful. Join over 170k+ readers who start their month with the Todoist newsletter. Over 177,000 subscribers Substack's , and and . ? # 😅 March: Lost ### Time to read: 5 mins Naomi from Todoist Mar 01, 2025 2 Share ### "I must remember that". There are very few things that will cause more instant mental anguish and frustration than remembering something really important (or worse, getting a genius idea) and promptly losing it to the cognitive abyss. 🕳️ That feeling of desperately grasping at something in your mind that was immediately important but seems nowhere to be found. You might put your hands to your head, close your eyes or attempt to quieten anyone talking to you as if halting all mental input will give your lost thought the space to return. It often doesn't. When it was a really great idea or must-remember-task, and you fail to bring it back, that feeling of deep frustration is probably going to hound you for the rest of the day. Enter: a quick and easy capture system. The speed at which you can get thoughts out of your head and into a safe place is a huge aspect of being an organized person. Removing friction is key. If you need to unlock your device, open an app, navigate to a button or a project and start typing, then _gooooood_ luck trying to hold onto that thought while you do so. 😅 You're more likely to mutter _"I'll remember it later"_. But that’s wishful thinking. These little optimistic fictions we tell ourselves usually amount to nothing, and your precious idea vanishes forever. 🫠 * * * **Here are the solutions:** 1. **Pen and paper: The original quick capture** If analogue is your thing, a small notebook and pen can be the perfect low-tech capture system: * Pocket-sized notebooks require zero startup time – flip and write. * Field Notes, Moleskine pocket notebooks, or even a stack of index cards held together with a binder clip fit easily in pockets and bags. * Strategically place notepads where thoughts often strike: bedside, bathroom (yes, waterproof shower notepads exist!), kitchen, car dashboard, etc. 1. **Todoist: Designed for lightning-fast capture** I know that not everyone reading is a Todoist user (regardless, you're all welcome here 🫶), but for me, Todoist is hands down the world’s best quick capture system. It saves my sanity regularly and here’s how: * **Magical Q shortcut:** When on my computer, pressing Q from anywhere instantly opens the **Quick Add box**. Type and hit Enter – captured in seconds. * **Lock screen widget:** I added the Todoist widget to **my phone's lock screen**. One tap from a locked phone, type your thought, hit add – all without even unlocking your device. * **Watch-based voice input:** Tap the Todoist shortcut on **your smart watch** then use the built in voice add option to speak your tasks to your watch. * **Email forwarding:****Every Todoist account has a unique email address**. Forward anything to it, and the email becomes a task with the content attached – perfect for capturing action items from your inbox. * **Share extension:** Found something online you need to remember? Use the **browser share extension** to send it directly to Todoist in two taps. I keep the Todoist widget on my home screen, and the Q shortcut has become pure muscle memory. That tiny bit of habit-building has saved countless ideas from evaporating into thin air. 1. **A mind at ease** At the end of the day and during your **weekly review**, you can process these quick captures properly – assigning due dates, projects, and priorities when you have the mental space to do so. The critical part was catching the thought before it vanished. The most elaborate organization system is worthless if your thoughts never make it there in the first place. Quick capture isn't just about productivity – it's about mental peace. When you know you have a reliable way to catch every important thought, your mind can actually relax. 😌 _“Your brain is for having ideas, not holding them.”_ **David Allen, Getting Things Done** ### The J.O.T. Method 📝 If you’ve followed our sound advice and are now crying into your coffee thinking “I’m so overwhelmed I don’t even know where to start”, then please stop staring at your task list and go **watch this video** instead. 👀 Trust me, once your brain feels confident about it’s ability to tackle a few small tasks one by one, you can begin to tackle the bigger stuff like task management. But for now, take it easy tiger. Here's to normalising a mind at ease and never **losing** a brilliant idea again, **Naomi (🙋♀️human, not AI) and the Todoist team** * * * ### Psst… Do you use Todoist with your team? Get in touch 💌 If you and your team use Todoist to plan your workload, then I want to talk to you. We’re looking to speak to as many professionals as we can who rely on Todoist to get work done together. Just hit reply to this email to contact me directly. I’d be so excited to talk with you about what your team are finding useful. * * * #### to Todoist Newsletter By Naomi · Launched a year ago Productivity inspiration and tactical advice that’s actually useful. Join over 170k+ readers who start their month with the Todoist newsletter. Substack's , and and . [2 Likes](https://todoist.substack.com/p/11837470_mar
Vídeos
Ver vídeo
Assistir no YouTube ↗Ver vídeo
Assistir no YouTube ↗adding more tools makes your agent worse
adding more tools makes your agent worse - OpenClaws and smarter routing for token efficiency. Copy the configuration. Run your 24/7 team. An independent AI agent community by Josh Davis. Over 38,000 subscribers Substack's , and and . ?### most openclaw agents don’t fail because they’re missing capability OpenClaw Mar 20, 2026 30 3 6 Share i 100% didn’t believe that at first. i kept adding tools thinking i was making the system stronger. it looked more powerful. it felt more complete. but every time i added something new, performance slipped. slower runs weirder decisions higher cost less trust i blamed the model. that was wrong. This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber. ## the real problem every tool is a decision. and the model has to get that decision right. every step. not just: “can this be done” but: * should i use a tool * which one * why this one over the others * what inputs * when to stop this is where things break. as tool count increases, selection accuracy drops and inefficiency rises you are not just adding capability. you are increasing the chance of being wrong. ## the failure you don’t see it doesn’t crash. it drifts. * it picks a “good enough” tool instead of the right one * it chains unnecessary steps * it ignores better paths * it behaves differently across runs you end up with something that looks smart… but you don’t trust it. ## the part most people miss tools are not just actions. they are context. every tool adds: * description * parameters * structure that all gets injected into the prompt. and models don’t process everything equally. they prioritize, skip, and compress. so what happens: important tools get buried irrelevant tools get chosen decisions get noisier this is not randomness. it’s overload. ## the punchline your agent isn’t failing because it’s weak. it’s failing because it has too many choices. ## the moment it clicked i stripped a workflow down. didn’t improve it. didn’t upgrade the model. just removed tools. and everything got better. faster cleaner predictable this is not just anecdotal. real systems are seeing the same thing. one team removed ~80% of their tools and saw: * success rate: 80% to 100% * execution time: ~275s to ~77s * tokens: ~102k to ~61k * steps: ~12 to roughly ~7 they didn’t add intelligence. they removed noise. ## the money part (this is what actually matters) this is the difference between: a workflow that costs cents and one that quietly burns dollars every run same task. different tool exposure. one makes money. one leaks it. ## what’s actually happening three things compound fast: ### 1. decision overload more tools = more comparisons the model spends more time deciding than executing ### 2. context dilution more tools = more noise relevant options get buried ### 3. path explosion more tools = more possible chains more chains = more failure paths this is why performance degrades as systems scale in complexity not because models are weak because systems get messy ## the hidden cost nobody tracks bad tool paths don’t always fail. they just become inefficient. more steps more retries more tokens and you don’t notice. until the bill shows up. ## the deeper problem tool usage has three failure points: * deciding if a tool is needed * selecting the correct one * using it correctly all three get worse as tool count increases. and most systems do nothing to reduce that burden. they just keep adding more. ## what actually works not more tools. less exposure. systems that perform well: * limit visible tools per task * merge similar tools * hide tools unless needed * replace repeatable paths with deterministic logic there is active work showing that filtering tools before exposure improves both accuracy and efficiency in large tool environments the direction is clear. not expansion. compression. ## use this immediately run this on any workflow you have. ``` act as a systems optimizer for agent workflows. goal: reduce tool-induced failure without reducing the outcome. input: - workflow goal - current tools - tool descriptions tasks: 1. find overlapping tools 2. find tools that create ambiguity 3. find tools that are rarely needed 4. identify tools that should be conditional 5. identify tools that should be replaced with deterministic functions 6. reduce to the minimum viable toolset 7. rank remaining tools by: - necessity - risk - likelihood of incorrect selection output: - failure risks - tools to remove - tools to merge - simplified architecture - why this improves performance ``` ## the rule i use now before adding anything: * does this really change the outcome * is there already overlap * will this create confusion * can this be merged * what happens if it’s used incorrectly if the answer isn’t obvious it doesn’t get added ## a better starting point start with only three buckets: * retrieval * transformation * action force everything into one. if it doesn’t fit cleanly you probably don’t need it yet ## the uncomfortable truth most people are building agents like feature lists. more integrations more connectors more actions it feels like progress. it isn’t. real systems don’t fail because they lack features. they fail because they can’t choose correctly. and every tool you add makes that harder. ## final point the best agents i’ve seen are not impressive. they’re major boring. they do **one thing.** with very few options. and they work every time. that’s what actually wins. This Substack is reader-supported. To receive new posts and support my work, consider becoming a free or paid subscriber. 30 Likes∙ [6 ](https://substack.com/note/p-191603443/ ?utm_source=substack&utm_content=facepile-) 30 3 6 Share #### Comments ian kachadorian [Mar 20](https://openclawunboxed.com/p/adding-more-tools-makes-your-agent/comment/230730036 "Mar 20, 2026, 5:35 PM") Liked by OpenClaw less is more honestly I don't want these things invading my privacy [Like (1)](javascript:void(0))ReplyShare Don Southerton [Mar 22](https://openclawunboxed.com/p/adding-more-tools-makes-your-agent/comment/231504102 "Mar 22, 2026, 1:03 PM") Love your images... Josh, I'm pondering when to jump into more Claw. Claude co-work is now my go-to. [Like](javascript:void(0))ReplyShare 1 more comment... Top Latest [st
Essa skill do Claude Code faz designs IRRESISTÍVEIS
Essa skill do Claude Code faz designs IRRESISTÍVEIS ### Como transformar landing pages genéricas “feitas por AI” em designs bonitos e distintos Deborah Folloni fev 18, 2026 Eu descobri um negócio que tá mudando completamente o jeito que eu crio interfaces com AI, e esse negócio é a **skill de frontend design da Anthropic** (basicamente um playbook que ensina seu coding assistant a fazer designs bonitos de verdade) - que não à toa é uma das skills mais instaladas no mundo com mais de 73 mil instalações. A diferença entre usar e não usar essa skill é a diferença entre criar uma página genérica “cara de AI” e criar algo com **identidade visual forte e profissional**. Testei ela na prática e o antes/depois foi tão absurdo que eu precisei gravar um vídeo inteiro sobre isso. A diferença não é só meramente estética. Essa skill te ajuda a criar algo que tenha identidade, que seja memorável, e que não pareça mais uma landing page genérica / AI slop. Vou te mostrar exatamente como usar essa skill no seu projeto e transformar seus designs de “meh” para “UAU”. ## O que é uma “skill”? Antes de mais nada, se você ainda não tá familiarizado com o conceito de skills: pensa que uma skill é como se fosse um **playbook que sua AI vai ler quando precisar fazer alguma coisa**. Você pode ter inúmeras skills instaladas no seu coding assistant. E diferente do system prompt (que é lido toda vez que você envia um prompt), **a skill só vai ser carregada no contexto certo**. Por exemplo: se você disser no prompt que precisa fazer um design e o Claude Code vê que você tem uma skill de design instalada, aí ele vai ler a skill para entender como fazer design do jeito certo. Eu inclusive já mostrei por aqui a skill do Remotion (que ensina o Claude Code a fazer animações com React). É o mesmo princípio, só que nesse outro caso, o “playbook” ensinava o Claude code a criar animações com o Remotion. ## Sem skill vs. com skill Para vocês entenderem o impacto dessa skill, eu fiz um teste prático com uma landing page pro meu projeto. **Primeiro round: SEM a skill de frontend design** Peguei um arquivo markdown com toda a copy da página (seção hero, social proof, comparação com concorrentes, features, como funciona, depoimentos, FAQ, etc.) e pedi pro Claude Code criar a landing page. O resultado? Bem... funcionou. Tem todos os elementos que eu pedi. Mas: Deborah Folloni | Youtube * Layout genérico * Cores sem personalidade * Tipografia padrão * Zero identidade * **100% cara de AI** É aquela página que poderia ser literalmente de qualquer produto. Não tem alma. Não tem diferenciação. **Segundo round: COM a skill de frontend design** Aí eu instalei a skill de frontend design, criei alguns assets básicos da marca (logo dark, logo light, paleta de cores, tipografia), e pedi pro Claude Code melhorar o design usando a skill e aplicando os assets da minha marca, e o resultado foi **completamente diferente**: Deborah Folloni | Youtube * Design sofisticado e distinto * Cores da marca aplicadas estrategicamente * Tipografia com hierarquia clara (fonte marcante pros títulos, fonte legível pro corpo) * Micro-interações (hover effects, animações sutis) * **Identidade visual forte** ## Não se esqueça de criar uma identidade visual Uma coisa que, sem dúvidas, faz muita diferença é ter uma identidade visual básica para o seu projeto antes de usar essa skill. Você não precisa ser designer pra fazer isso! Eu já ensinei em uma outra edição da minha newsletter como criar essa mini identidade visual mesmo sem habilidade nenhuma de design. Deb GPT Como criar uma identidade visual premium usando IA (gastando zero reais) Deixa eu te contar uma coisa: esse é o tipo de trabalho que qualquer agência te cobraria fácil de 5 a 10 mil reais para fazer. E hoje eu vou te mostrar como fazer sozinha, sem saber absolutamente nada de design, usando ferramentas de IA que qualquer pessoa consegue usar… Read more 2 months ago · 177 likes · 10 comments · Deborah Folloni Você só precisa do básico: * **Logo** (com versão dark e light) * **Paleta de cores** (3-5 cores principais com seus códigos hex) * **Tipografia** (uma fonte pra títulos, outra pro corpo do texto) * **Conceito visual** (ex: montanhas, tecnologia, minimalismo) Eu fiz isso pro meu projeto em menos de 30 minutos usando ferramentas de AI. E o resultado foi suficiente pra dar uma identidade forte pra página. **Por que isso importa?** Porque sem essa identidade básica, ainda que essa skill de frontend design deixe seu projeto BONITO, ele não vai ficar DISTINTO - ou seja, seu projeto vai ficar parecido com todo mundo. A skill potencializa sua identidade, mas você precisa ter uma identidade para potencializar 😉 ## Como instalar e usar a skill A instalação é super simples. Existem vários diretórios de skills por aí, mas eu gosto muito do **skills.sh** (feito pela Vercel). Ele também mostra quantas instalações cada skill teve - o que ajuda a identificar quais skills realmente valem a pena. skills.sh ### Instalação Você tem duas opções: **Opção 1:** Rodar no terminal ``` npx skills add https://github.com/anthropics/skills --skill frontend-design ``` **Opção 2:** Colar esse prompt no Claude Code e pedir pra ele instalar pra você (mais prático) ``` "Help me install this skill npx skills add https://github.com/anthropics/skills --skill frontend-design" ``` Depois da instalação, você precisa começar uma conversa nova no Claude Code pra ele detectar a skill. ### Como usar O workflow que eu usei: 1. **Prepare seus assets:** Coloque no projeto seus logos (dark + light), uma imagem de fundo/ assets da sua marca se tiver, e tenha em mãos os códigos das suas cores e nomes das fontes da sua marca. 2. **Monte seu prompt:** Diga pro Claude Code que você quer melhorar o design usando a skill de frontend design, mencione onde estão os assets da sua marca dentro do seu projeto (ex: pasta /public/assets): ``` "help me improve the design of this landing page using frontend design skill. Use font [title font name] for titles and [body font name] for body. Use #XXXXXX, #XXXXXX and #XXXXXX as brand colours for this project. Use assets in folder public/assets" ``` 1. **Deixe a skill trabalhar:** O Claude Code vai ler as melhores práticas da skill e aplicar no seu design ## Por que isso importa pra você Dizem que *“taste is the new moat”* e num mundo onde todos os designs parecem iguais, isso não poderia ser mais verdade. Você pode ter o melhor produto do mundo, mas se sua landing page parece “feita por AI” e sem identidade, você perde credibilidade. As pessoas julgam pela capa, sim. A boa notícia é que essa skill, você não precisa gastar horas aprendendo design e ter que usar ferramentas mais complexas como o Figma. Você só precisa: * Criar uma mini identidade visual básica * Instalar a skill * Fazer prompts que eu ensinei Assim, você terá um design profissional, distinto, e que não grita “feito AI”. --- **Quer ver todo esse processo na prática?** Eu gravei um vídeo completo mostrando o antes e depois, o passo a passo da instalação e os resultados completos! Até a próxima edição da DebGPT! 💜 #### Discussão sobre este post Jesse santos [Feb 18](https://dfolloni.substack.com/p/essa-skill-do-claude-code-faz-designs/comment/216250263 "Feb 18, 2026, 3:31 PM") Cada dia que passa eu vejo que todo mundo que utiliza IA de forma mais completa utiliza o Claude. Estou sou novato no mundo da IA e achava que o GPT era o ápice mas vendo todas as funcionalidades e soluções que vocês têm no Claude estou achando incrível. Compartilhar 1 resposta [Feb 18](https://dfolloni.substack.com/p/essa-skill-do-claude-code-faz-designs/comment/216243655 "Feb 18, 2026, 3:17 PM") Adorei, já vou aplicar, hj ta assim: www.agrozappi.vom.br ResponderCompartilhar 4 comentários a mais... Nenhuma publicação ### Pronto para mais? © 2026 Deborah · Privacidade ∙ Termos ∙ Aviso de coleta Comece seu SubstackObtenha o App Substack é o lar da grande cultura
Prompts
"Help me install this skill npx skills add https://github.com/anthropics/skills --skill frontend-design"
"help me improve the design of this landing page using frontend design skill. Use font [title font name] for titles and [body font name] for body. Use #XXXXXX, #XXXXXX and #XXXXXX as brand colours for this project. Use assets in folder public/assets"
👋 May: We’re leaving…
👋 May: We’re leaving… - by Naomi from Todoistseful. Join over 170k+ readers who start their month with the Todoist newsletter. Over 177,000 subscribers Substack's , and and . ? # 👋 May: We’re leaving… ### Time to read: 3 mins Naomi from Todoist May 01, 2025 12 1 2 Share ### Step away from the app store. I’ve been spending an embarrassing amount of time researching fitness trackers. The Oura ring to be specific. Not because my Apple Watch is broken. Not because I need advanced features. But because I haven’t exercised in over a month (😬), I am somehow convincing myself that a new device and fitness app will fix that. _“I’m more likely to run if I have more data on fancy charts.”_ 🙄 Sound familiar? This is precisely the little voice that pops up when Todoist (or your productivity partner of choice) isn't "working" for you: _“Maybe I just need a fresh start.”_ _“Their Notion dashboard is so pretty!“_ _“A paper planner would make me feel more in control.“_ It's tempting to think the solution is a better tool. But most of the time, the tool isn't the problem. Here's the uncomfortable truth… You're probably not behind because your app doesn’t cut it. You probably just feel behind because you're human, life is complex, and motivation may be hard to come by right now. Switching systems can feel like progress (I’ve done this more often than I’d like to admit). But don’t be fooled! It often just delays doing the things you actually need to do and gives you a false sense of accomplishment (while the actual tasks pile up). * * * **If your setup feels off, before you go app-shopping, consider asking:** * Do I actually need a new tool? Or do I just need to drastically simplify the one I have? * Am I using this tool intentionally? Or am I just dumping everything into it and never revisiting with a plan? (We can help with **the planning part**by the way 😉) * Would 15 minutes of simple maintenance (deleting those 253 overdue tasks from 2023) help me more than a complete system overhaul? Let's be honest — a new app won't magically make us do the thing we've been avoiding for weeks. 😅 Often, the solution isn't a new tool but just sitting down and getting started. No app in the world can do that part for us… _[AI enters the room]_ Just wake me up when AI can go on a run for me. 🙃 **Open your Todoist** **“It is never the tool that decides. It's the hands-and the heart-of the one who wields it.”** Kevin Sands, The Blackthorn Key ### A Question 🤖 A recent **Harvard Business Review article**made the rounds at Todoist for its insightful look at how people are _actually_ using AI in 2025. It sparked a broader conversation about how we’re all using it in our daily lives. Soon, we’ll share more of our thoughts but in the meantime, we’d love to hear yours. How do you use AI day-to-day? Let us know in **this short survey**. * * * ### House Move 🚛🍕 We interrupt your regular reading pleasure to let you know we're packing up our little newsletter suitcase and moving it over to Substack. _Why?_ Because Mailchimp feels like sending emails from a fax machine, and Substack lets us build more of a community with you. There’s a comment box to engage with, and the emails are way easier to share and revisit without having to riffle through your inbox. You don’t need to do a thing (and to be honest, you probably won’t even notice.) You’ll keep getting the same productivity ponderings on the 1st day of every month. Just from a cozier, more conversational spot. **👉 If you're happy with that, carry on with your day as usual** **👉 If you'd rather not make the move with us, there's a guilt-free link below** If you choose to stick around, thank you. 🙇♀️ See you on 1st June! **Naomi (🫶 real human) & the Todoist team** 12 Likes∙ [2 ](https://substack.com/note/p-163690179/ ?utm_source=substack&utm_content=facepile-) 12 1 2 Share #### Comments Tina Townsend [May 18, 2025](https://todoist.substack.com/p/12824218_may-2025-newsletter/comment/118229298 "May 18, 2025, 9:20 PM") Liked by Naomi from Todoist I will follow you anywhere! TODOIST has changed my life. I am going through several life changes and this app has helped me so much. I do need to make more of a plan and I am working on setting things up differently. I am 58 and I am going back to college to learn more about my passions. It might turn into a side hustle or a whole new gig who knows! I try to watch all of your videos. The Mind Dump was so helpful. I have a notebook I do a Mind Dump then the ones that I feel are important I put in TODOIST! Thank you for all you do! Don't ever stop! Tina R Townsend 🩷 [Like (5)](javascript:void(0))Reply[Share](https://todoist.substack.com/p/12824218_may-2025-newslet
📈 January: Set Yourself Up
📈 January: Set Yourself Up ### Four realistic tips for 2026, with a dose of honest New Year energy Naomi from Todoist Jan 01, 2026 #### Happy 2026. 🥂 I don’t plan to waste your time with saccharine “New Year, New Me” content. Fresh starts are phenomenal. I love them. But I also want you to succeed at whatever you want your 2026 to look like. And most “New Year, New Me” advice falls short of empowering you to make that happen. So here are four things that I know to be true and that I believe will propel you forward this year, way faster than any “first month free” gym membership or green powder subscription will. It’s the scaffolding I’m using for my own 2026 plans. Without further ado … --- ### 💆♀️ Take the whole of January Every year around this time, most of the world seems to agree that the first of January is the moment you’re supposed to be ready. Your goals should be locked in, your habits decided, and your game plan set. Off you go. I’ve never quite bought into that. For as long as I can remember, I’ve taken the whole of January to figure things out. Not because I’m indecisive, but because I’ve learned (the hard way) that urgency is a terrible environment for good decisions. One rushed day of planning rarely beats a few weeks of taking stock, adjusting, and setting myself up properly. Honestly? Starting in February is underrated. The gyms are quieter. Most people have already forgotten what they promised themselves. And there’s something about the rebellion of improving my life on my own timeline that appeals to me. So take your sweet time. One month of unhurried preparation, habit forming, and tapping into what you *really* want, will set you up for eleven months of sustained energy and focus better than a first-of-January outburst ever will. --- ### 🥴 Stop overloading your brain If there’s one thing that reliably derails any good intention, it’s overwhelm. * The thoughts that pop up while brushing your teeth. * The idea that arrives mid-conversation. * The reminder that appears just as you’re about to fall asleep. You feel the urge to *write that down* – but it ends up on the back of an envelope, buried in Notes, or lost to the abyss of “I’ll remember this later.” (You won’t). Over time, those loose ends pile up as low-level background stress. Your brain keeps nudging you because it doesn’t trust that anything important is being looked after. Feeling organized has very little to do with color-coded systems or perfect plans. It’s basically all about trust. And your brain loves to trust a single source of truth. So pick **one** place to park those thoughts. And use it every day. It can be anywhere. A single notepad will do. As you’ll know, for me, it’s Todoist. And even more so now that we’ve released the feature, Ramble. Speaking messy thoughts and watching Ramble magically transform them into tasks is the catharsis I didn’t know I needed. I recently posted an impromptu video on LinkedIn of how I use it to plan the weekend’s meals; it’s a kinda show-don’t-tell feature. Just pick one place to offload your mental chatter, that’s the takeaway. And stick with it. Once you start to trust that system, you’ll reach a greater level of peace. --- ### ☕ Make time to plan (and make it enjoyable) Getting things out of your head is only half the story. No matter where you put them, if they stay there untouched, they get stagnant and weigh you down. This is where planning comes in. And yes, I’m talking about the weekly review/reset. Again. I know you might see this as boring, but trust me: If there’s one thing I’ve learned in my almost 40 years on this planet, the boring, repetitive stuff is where the compounding magic is. For me, a weekly reset is a slightly indulgent time. A cup of something delicious. A bit of space. A chance to look at what’s coming up and deliberately decide what to do about it. A weekly reset keeps your task list fresh. Without it, you run the risk of stagnation, but we want that list to flow. It’s the habit that transforms you into the colleague who submits the document two days early, or the parent in the group chat who reminds everyone else it’s silly hair day on Wednesday. It’s about being proactive about your time, instead of reactive. A weekly reset gives you 52 fresh starts a year. Imagine what you could achieve with that level of intention. If you want to know more about what’s involved in my own weekly reset, I’ve made a video about that on Todoist’s YouTube channel. --- ### 🎨 You don’t need to change your life to change your life Last year, I often fell into a trap I know many of you will recognize. I kept waiting for the perfect conditions to make progress. Long, uninterrupted afternoons. Clear desks. Empty houses. Big, sprawling stretches of time where I could finally *get into it*. They rarely arrived. What did happen were micro-moments. Ten minutes here. Twenty minutes there. Small pockets of time that I dismissed as “not enough” until I started using them. * I learned a new painting technique in 10 minutes. * I fixed a broken kitchen shelf in 20. * I taught myself the basics of “Carol of the Bells” on piano while a sticky toffee pudding cooked in the oven. Many things don’t take as long as we think they do. And most progress doesn’t require a life overhaul, just a willingness to start where you are, imperfect conditions and all. So the time you find yourself with a little pocket of free time, take a micro-moment to do something you’ll feel smug about later. It does wonders for your self-esteem. --- To recap: * Plan your year throughout January * Pick one place to get everything out of your head * Plan your time weekly * Take advantage of micro-moments Implement one or all of these, and I have faith that you can tackle anything (both cheerful and challenging) this year throws your way. "Little by little, a little becomes a lot." Tanzanian proverb ### 💡 Todoist Tip According to last month’s reader survey (thank you for your kind responses 🫶), you want us to continue with the chatty nature of the newsletters, and you’d like me to bolt on a tip on how to get the most out of your Todoist. Your wish is my command; 🧞♀️ consider this a new recurring section. Want to get faster at navigating your Todoist from a keyboard? There’s an expanding variety of keyboard shortcuts on the Todoist desktop apps. Open any project or view in Todoist, then select `?` on macOS, web, and Windows to see the full list. Technically, `?` is the shortcut to learn the rest of the shortcuts. 😏 A handy one to have in your back pocket. --- And just like that, a new year begins. I’d insert an inspirational parting message, but really, only you know what you want to get out of your 2026. And we’re cheering you on. Naomi (🙋♀️ real human) and the Todoist team --- to get productivity inspiration & tactical advice that’s actually useful (and free). Sent on the 1st of each month. 💌 #### Jennifer Trumbore [3d](https://todoist.substack.com/p/january-set-yourself-up/comment/193892844 "Jan 1, 2026, 1:35 PM") Thank you for the thoughts - very practical and helpful! I wonder if there is a shortcut on mobile devices to get to Ramble quicker? I always think up the best things to do when I'm drivi
Vídeos
Ver vídeo
Assistir no YouTube ↗AI tutor agents, omnimodal video models, LTX-2 updates, long-term memory, video faceswap: AI NEWS
[AI Search](/) AI Search AI tutor agents, omnimodal video models, LTX-2 updates, long-term memory, video faceswap: AI NEWS 0:00 Current time: 0:00 / Total time: -35:40 -35:40 ## AI tutor agents, omnimodal video models, LTX-2 updates, long-term memory, video faceswap: AI NEWS Welcome to the AI Search podcast. Here are the top highlights in AI this week AI Search Jan 13, 2026 HUGE AI NEWS: LTX-2, UniVideo, SimpleMem, HY-MT, NeoVerse & more #ai #ainews #aitools #aivideo Note that most of the examples are visual. See my Youtube video for the best experience. AI Search AI news, science, & research. Stay up to date with AI every week. AI news, science, & research. Stay up to date with AI every week. Substack App Apple Podcasts Spotify RSS Feed Appears in episode AI Search Recent Episodes [Realtime AI video, -level AI agents, realtime TTS, new open music generator, new video animator: AI NEWS](https://aisearch.substack.com/p/realtime-ai-video- -level-ai-agents) • AI Search DeepSeek breakthrough, realtime upscaler, realtime 3D worlds, new top 3D generator, Qwen Image 2512: AI NEWS • AI Search New open Nano Banana, AI plays any video game, new top open source models, long AI videos: AI NEWS • AI Search Insane 3D models, realtime AI video, new #1 open model, realtime AI worlds, Gemini 3 Flash: AI NEWS • AI Search [GPT 5.
Realtime AI video games, Moltbook, agent swarms, AI Earth, open video models: AI NEWS
[AI Search](/) AI Search Realtime AI video games, Moltbook, agent swarms, AI Earth, open video models: AI NEWS 0:00 Current time: 0:00 / Total time: -40:52 -40:52 ## Realtime AI video games, Moltbook, agent swarms, AI Earth, open video models: AI NEWS Welcome to the AI Search podcast. Here are the top highlights in AI this week AI Search Feb 07, 2026 HUGE AI NEWS: Google Genie 3, Lingbot World, Moltbook, Clawdbot, Kimi K2.5, Qwen Max #ai #ainews #aitools #aivideo Note that most of the examples are visual. See my Youtube video for the best experience. AI Search AI news, science, & research. Stay up to date with AI every week. AI news, science, & research. Stay up to date with AI every week. Substack App Apple Podcasts Spotify RSS Feed Appears in episode AI Search Recent Episodes Realtime AI voices, AI livestreamers, Blender 3D agents, realtime worlds, new top OCR: AI NEWS • AI Search [Realtime AI video, -level AI agents, realtime TTS, new open music generator, new video animator: AI NEWS](https://aisearch.substack.com/p/realtime-ai-video- -level-ai-agents) • AI Search AI tutor agents, omnimodal video models, LTX-2 updates, long-term memory, video faceswap: AI NEWS • AI Search DeepSeek breakthrough, realtime upscaler, realtime 3D worlds, new top 3D generator, Qwen Image 2512: AI NEWS • AI Search [New
My Deep Learning Math Workbook — Now Interactive
[AI by Hand ✍️](/)— Now Interactive Happy to share something new: an interactive edition of my popular Deep Learning Math Workbook. This is a proof-of-concept prototype with the first 15 exercises of Chapter 1: Dot Product. You fill in the blanks, pick from multiple choices, and check your answers — all in your browser. 👉 Give it a try Let me know what you think. I still believe pen and paper is the best way to learn math. There’s something about the focus and concentration that a blank page demands. But many of you have asked for an interactive option, and I think there’s real value in being able to test yourself and get immediate feedback. ~ Prof. Tom Yeh #### Interactive format is more engaging 👏🏻✅ Done it all correct at 1️⃣ shot without hallucinations 🥱 🤭 No posts ### ?
Beyond the “Gradient Highway”: How Attention Residuals Fix the Hidden Crisis of Deep LLMs
Beyond the “Gradient Highway”: How Attention Residuals Fix the Hidden Crisis of Deep LLMs # Agentic AI“Gradient Highway”: How Attention Residuals Fix the Hidden Crisis of Deep LLMs ### Some key takeaways from recent paper from open source model Kimi’s research Ken Huang Mar 18, 2026 ∙ Paid 14 1 Share Sometimes, most advanced research does not need a PhD; the X post about Kimi’s new AI architecture, allegedly created by a 17‑year‑old and praised by Elon Musk, is a vivid example of how exceptional talent, open research, and plentiful compute can let independent young researchers push state‑of‑the‑art AI infrastructure forward without formal credentials, delivering “drop‑in” architectures with meaningful compute gains at minimal extra latency and proving that breakthrough ideas increasingly come from those who move fastest, not those with the longest academic résumés, and our AI researchers at DistributedApps.ai have analyzed the underlying work and summarized the key takeaways— to our paid tier to know more. 1. Introduction: The Amnesia of Depth For years, the “gradient highway” has been the structural backbone of deep learning. Standard residual connections serve as a fast lane, allowing information to bypass complex transformations via identity mappings. But this high-speed travel carries a hidden cost. Imagine a highway where every town passed adds new cargo to a truck. By the time the vehicle has traveled thousands of miles—or passed through hundreds of neural layers—the original items from the start of the trip are buried under a mountain of new weight. In modern Large Language Models (LLMs) using PreNorm architectures, this manifests as a form of architectural “amnesia” or dilution. Because standard residuals aggregate information using fixed unit weights, each individual layer’s contribution is progressively washed out as the model grows deeper. Attention Residuals (AttnRes), an innovation recently detailed by the Kimi team at Moonshot AI, introduces a “selective memory” upgrade. By moving away from blind accumulation, AttnRes allows each layer to perform content-aware retrieval across the model’s entire history. 1. The PreNorm Paradox: Why More Layers Don’t Always Mean Better Features While depth is intended to build increasingly sophisticated features, modern PreNorm architectures suffer from a technical limitation: hidden-state magnitudes grow as O(L) with depth. As these magnitudes expand, the relative influence of any single new layer—and its ability to impact the final output—shrinks. Early-layer information is effectively “buried,” and the model loses the capacity to retrieve specific representations from its own past. Standard residuals trap the model in a “fixed unit weight” strategy where every layer is treated with equal importance, regardless of its utility. As the research team observes: “Residuals also play a second role that has received less attention... residuals define how information aggregates across depth. Unlike sequence mixing and expert routing, which now employ learnable input-dependent weighting, this depth-wise aggregation remains governed by fixed unit weights.” 1. Takeaway #1: Replacing Addition with Selection (Softmax over Depth) The central intellectual pivot of AttnRes is the Duality of Time and Depth. Just as the Transformer revolution replaced the sequential recurrence of RNNs with attention across time (the sequence), AttnRes replaces the additive recurrence of residuals with attention across depth. ## Continue reading this post for free, courtesy of Ken Huang. Claim my free post [Or purchase a paid subscription.](https://kenhuangus.substack.com/ ?simple=true& =https%3A%2F%2Fkenhuangus.substack.com%2Fp%2Fbeyond-the-gradient-highway-how-attention&utm_source=paywall&utm_medium=web&utm_content=191286591&just_signed_
My most-used phrase with AI is “ask me one question at a time.”
My most-used phrase with AI is “ask me one question at a time.” ### Because chat LLMs aren’t very conversational Tal Raviv Jan 15, 2026 Wispr (my speech to text app) says my most-used phrase with AI is “ask me one question at a time.” It got me to reflect on why I say this so often. I realized that chat LLMs aren’t very conversational. (Think about the last time ChatGPT answered with a wall of text.) When I collaborate with an LLM and it answers with a novel, it reminds me of that junior person who gets a task and runs off for two weeks in a cave, working super hard on it. They come back and do a velvet reveal.... that’s completely misaligned. The longer the LLM’s response, the more assumptions it’s making. For example, imagine if an LLM replies with 8 questions for you to answer. If your answer to the first question isn’t what it expected, the other 7 become irrelevant. **So, when I say “ask me one question at a time” I’m giving AI a chance to align and recalibrate after each question.** This also extracts more context out of me that I wouldn’t have articulated upfront, and gives me tons of chances to apply my taste along the way. Opus responded with a single spot-on question, then paused for my response. Without that phrase, Opus printed four mostly irrelevant options (each with pros and cons), plus a section it titled “my instinct” and ended with 3 questions for me, all in the same breath. Whether I’m writing a spec, drafting a research plan, analyzing data, working on positioning, or just drafting an update: I get way better results, way faster, when I make it a conversation. It’s the same thing we already know about collaborating with humans: frequent alignment beats cranking away in isolation. #### Sandra Herz [Jan 15](https://www.talraviv.co/p/wispr-says-my-most-used-phrases-with/comment/200226379 "Jan 15, 2026, 4:29 PM") Plus we will all get more aware of wasting Tokens when the comfortable flat fee system comes to an end (and I am pretty sure it will, sooner or later). ReplyShare Rebecca Spitzer [Jan 15](https://www.talraviv.co/p/wispr-says-my-most-used-phrases-with/comment/200146059 "Jan 15, 2026, 1:37 PM") I was just saying to my husband last night that this is one of my favorites!! I hate when it asks me 25 questions and I have to answer at once. I love the pattern Claude / Claude Code
Machine Learning Cheatsheets
Machine Learning Cheatsheets ### Covering core topics in Artificial Intelligence, Machine Learning, Deep Learning and Generative AI. Rohit Kumar Tiwari May 18, 2025 I’m super excited to share something I have collated recently: **AI‑ML‑Cheatsheets**, a GitHub repository. These cheatsheets are designed by Standford University to help students, developers, and researchers quickly recall important concepts and formulas. Each cheatsheet combines concise explanations, intuitive diagrams, and the key equations you’ll want to keep top of mind. I’ve organized everything folder wise by topic, so you can jump straight to what you need: --- ## **1. Artificial Intelligence** --- ## **2. Transformers and LLMs** --- ## **3. Deep Learning** --- ## **4. Machine Learning** --- ## **5. Probabilities and Statistics** --- ## **6. Algebra and Calculus** --- ## Check It Out! 🔗 **GitHub:** Feel free to explore, star, and clone. #### Happy learning! --- *Liked this article?* Make sure to 💜 click the like button. Feedback or addition? Make sure to 💬 comment. *Know someone that would find this helpful?* Make sure to 🔁 . ## **Get in touch** You can find me on LinkedIn | YouTube | GitHub | X Book an Appointment: Topmate If you wish to make a request on particular topic you would like to read, you can send me an email t
Multi-Agent Blog Generator
Multi-Agent Generator ### Generate high quality content with support for OpenAI, Gemini, Claude, and Grok. Rohit Kumar Tiwari Jul 06, 2025 I have been experimenting with different AI agent frameworks over the past few weeks. There are a lot of them now: LangGraph, CrewAI, Agno, Semantic Kernel, OpenAI Agents SDK, and more. Each of them has its own pros and cons but I really liked Agno Framework because its quite beginner friendly. So I used it to build something useful → a **Multi-Agent Generator**. The Multi-Agent Generator uses multiple AI agents that work together to generate professional, well-structured, and well-researched posts. Each agent handles a specific task. This makes the writing process more reliable. 1. **Search Agent**: Handles intelligent web searching and source gathering. 2. **Scraper Agent**: Extracts and processes article content. 3. **Writer Agent**: Generates the post based on research. ## Key Features * **Multi-Agent Workflow** Orchestrates research, content extraction, and writing using specialized agents. * **Provider Flexibility** Easily switch between OpenAI, Gemini, Claude, and Grok models. * **Research-Driven Content** Finds, evaluates, and cites trustworthy sources to back up claims. * **Content Scraping** Extracts and summarizes article content for reference. * **Gradio UI** Simple web interface for generating and viewing posts. ## How it works * Select your preferred LLM provider and model. * Enter your API key. * Choose an example topic or enter your own. * Click Generate to create a post. --- ### Clone the repository and start exploring! 👉 **Code** → 🚀 **Launch App** → --- #### ***Missed a recent post?*** **Catch up here 👇** 1. How RAG Enhances LLMs: A Step-by-Step Exploration 2. A Visual Guide to AI Agents 3. A Visual Guide to Agentic RAG --- *Liked this article?* Make sure to 💜 click the like button. Feedback or addition? Make sure to 💬 comment. *Know someone that would find this helpful?* Make sure to 🔁 . ## **Get in touch** You can find me on LinkedIn | YouTube | GitHub | X Book an Appointment: Topmate If you wish to make a request on particular topic you would like to read, you can send me an email to [analyticalrohit.connect@gmail.com](mailto:analyticalrohit.connect@gmail.com) --- Thanks for reading AwesomeNeuron! for free to receive new posts and support my work. #### [Rohit Kumar Tiwari](https://substack.com/profile/32
7 AI tools to help in your Work
7 AI tools to help in your Work ### These tools make you 10x more productive. Rohit Kumar Tiwari Aug 03, 2025 *“The biggest mistake you can make right now is ignoring AI tools.”* They're real, they're useful, and they're changing how people work. The people using them are getting more done with less effort and time. It’s not about replacing your skills. It’s about saving time and working smarter!! I also use AI tools every day. What I use them for: * Summarizing long articles * Brainstorming ideas * Writing emails faster * Breaking down research papers * Writing and debugging code Here are 7 tools that actually help: ### 1. ChatGPT (Brainstorming & Writing) Use it when you're stuck or need a second brain. It helps you: * Come up with ideas * Write emails, posts, or reports * Translate text or explain things clearly * Solve logic problems or tough questions ### 2. Perplexity (Smart Web Search) Think of it as a search engine that gives you answers. * It pulls info from the web and gives you a short, clear response * Shows sources so you can check for yourself * Good for quick research or fact-checking ### 3. Replit (Code and Build Apps) If you want to build something, but coding feels slow, use Replit. * Write what you want in plain English * It turns your ideas into working code * Fixes basic errors automatically * Helps you build and test apps fast ### 4. Windsurf (AI Help for Coders) This one helps you write better code faster. * Gives smart suggestions while you type * Can write full functions based on what you describe * Reduces time spent fixing bugs or writing boilerplate code ### 5. Grok (Deep Research) Perfect for complex deep research. * Reads long PDFs or documents so you don’t have to * Can search across multiple sources at once * Helps you break down complicated topics or papers ### 6. Gamma AI (Make Slides Fast) Making presentations is boring specially for tech folks. Gamma makes it faster and easier. * You type in your topic, and it builds a slide deck * Uses clean templates and auto-writes the content * Supports sharing and edits for team projects ### 7. Lovable (Build Websites with AI) No coding needed. Just describe your site, and Lovable builds it. * Create full websites from a short description * Works with GitHub if you want to add your own code * Great for portfolios, landing pages, or small business sites --- > Wait, did I miss your favorite tool? Tell me in the comments. I would love to hear about it. --- #### ***Missed a recent post?*** **Catch up here 👇** 1. A Guide to Becoming a 10x Vibe Coder 2. [Build your own Deep Researcher Agent](https://awesomeneuron.substack.com/p/multi-agent- -generator) 3. A Visual Guide to Agentic RAG 4. A Visual Guide to AI Agents --- *Liked this article?* Make sure to 💜 click the like button. Feedback or addition? Make sure to 💬 comment. *Know someone that would find this helpful?* Make sure to 🔁 . ## **Get in touch** You can find me on LinkedIn | YouTube | GitHub | X Book an Appointment: Topmate If you wish to make a request on particular topic you would like to read, you can send me an email to [analyticalrohit.connect@gmail.com](mailto:analyticalrohit.connect@g
7 Patterns for Smarter Multi-Agent Systems
7 Patterns for Smarter Multi-Agent Systems ### How multiple AI agents can work together as a team. Rohit Kumar Tiwari Nov 16, 2025 In A Visual Guide to AI Agent we saw how AI agents work, key agentic patterns, benefits and challenges they bring. Today, we will look at 7 key patterns for building multi agent systems. These patterns define how agents collaborate, share results, and manage their work. Choosing the right pattern can improve coordination, efficiency, and scalability. In simple terms, multi agent systems are teams of AI agents working together on a common goal. ## Why do we need Multi-Agent Systems? Think of a football team. No single player can do everything. You have: * Goalkeeper * Defenders * Midfielders * Forwards Each player has one main skill. They communicate and work together to win. If everyone tried to play every role at once, the team would break down. Multi Agent Systems work the same way. When a task is too big or too complex for one agent, we split the work among multiple agents. Each agent focuses on one type of job. One might search for information. Another might make decisions. Another might act on those decisions. By working together they can handle tasks that are difficult for a single agent alone. So multi agent systems are simply teams of AI agents working together, just like players on a football field. ## Key Patterns ### 1. Parallel Pattern * **How it works:** Several agents work at the same time on different parts of the same task, then combine results. * **When to use:** For large or complex problems that can be split into smaller independent parts. * **Example:** Multiple agents analyze different data sets or sections of a report, then merge their outputs into one final result. ### 2. Sequential Pattern * **How it works:** One agent completes its part and passes the result to the , like a chain of steps. * **When to use:** For problems that need a clear order or a fixed sequence. * **Example:** One agent drafts content, another edits, and a third reviews. ### 3. Loop Pattern * **How it works:** Agents keep reviewing and refining their work based on feedback until the output meets a goal. * **When to use:** For tasks that need multiple rounds of revision or learning. * **Example:** An agent generates an answer, critiques it, adjusts, and repeats the cycle until it’s confident in the final version. ### 4. Router Pattern * **How it works:** A central agent (or router) decides which specialized agent should handle a given input. * **When to use:** When tasks vary by type or complexity. * **Example:** A router sends coding questions to a programming agent and writing questions to a language agent. ### 5. Aggregator Pattern * **How it works:** Multiple agents work independently and a main agent combines their outputs into one final answer. * **When to use:** When different viewpoints or skills need to be merged. * **Example:** Agents provide alternative summaries or analyses, and an aggregator selects the best parts. ### 6. Network Pattern * **How it works:** Agents communicate freely with each other, sharing data and ideas. * **When to use:** For open ended or research driven systems that benefit from knowledge exchange. * **Example:** Agents discuss hypotheses, share discoveries, and adjust their reasoning together. ### 7. Hierarchical Pattern * **How it works:** Higher level agents manage and coordinate lower level ones, similar to a manager directing a team. * **When to use:** For large, multi-layered systems or projects that need strong oversight. * **Example:** A manager agent plans strategy, lower lever agents handle sub-tasks execution. --- ## ***Missed a recent post?*** **Catch up here 👇** 1. The Art and Science Behind Context Engineering 2. Small Language Models are the Future of Agentic AI 3. A Guide to Becoming a 10x Vibe Coder 4. [Build your own Deep Researcher Agent](https://awesomeneuron.substack.com/p/multi-agent- -generator) 5. A Visual Guide to Agentic RAG --- *Liked this article?* Make sure to 💜 click the like button. *Know someone that would find this helpful?* Make sure to 🔁 . ## **Get in touch** You can find me on LinkedIn | GitHub | YouTube | X | Topmate --- Thanks for reading [AwesomeNeuron](https://awesomeneuron.s
Is Venezuela’s Oil Worth the Hassle?
[Uncharted Territories](/)? Tomas Pueyo Jan 13, 2026 ∙ Paid Trump claims that the abduction of Maduro is all about the oil: capturing it for US use, while keeping competitors from its benefits. But how realistic is it? This depends on how much oil can be extracted from Venezuela. Today, it’s ~1.1M barrels per day.*Source* A barrel of oil is currently worth about $60: But Venezuela’s oil is worse quality than most, so it sells for cheaper, ~$8 less as of today, or $52.1 Let’s assume that continues. But how much does it cost to extract a barrel of Orinoco oil and transport it and treat it to be sellable? So of these $52, about $23 are hard costs, and each barrel yields around $29 in profit. For 1.1M barrels per day, that’s ~$32M per day, or ~$11.7B per year. That’s how much money Venezuela could earn from its oil. Compare this to the current situation in the country: * The Venezuelan government’s last budget asked for ~$20B, or almost twice the revenue that can be obtained from oil production.2 * Venezuela owes ~$200B, or nearly 20 years worth of production! *Source* ConocoPhillips alone is owed $12B, about a year’s worth of all of Venezuela’s oil revenue! So either the US makes no money from Venezuela, or it plunders the oil and leaves Venezuelans to starve and die—won’t happen—, or it invests to decrease costs and increase production. Now, decreasing costs is very hard because, as we saw in the article, the oil is superheavy. As Nigel Harris said in this comment: > *In Saudi Arabia, all you have to do to extract the oil is drill a well, and control the resulting flow of oil and gas, which comes bursting out of the ground under its own geological pressure. Let the oil sit in a tank for a short while, and the gases bubble off (and are captured for use as fuels) and the water and sand settle out. The oil is ready to transport by pipeline and ship, and is of a quality readily handled by almost any fuel refinery in the world.* > > *The oil* [in the Orinoco Valley] *is extremely dense (heavier than water), extremely viscous (like pitch or molasses) and extremely dirty (over 5% sulfur and masses of metals like vanadium). The only deposit like this elsewhere in the world is Canada’s Athabasca oil sands.* > > *To extract the oil, you have to first pump large amounts of steam into the formation, to melt the hydrocarbons, then use electrical pumps at the surface or in the bottom of the well, up to a kilometer deep, to lift it to the surface. Once there, the “oil” is far too viscous to transport by pipeline or ship, and far too heavy and dirty for most refineries to tackle. So it is diluted by mixing with a much lighter crude oil, or the “condensate” liquids from a gas field, or refined naphtha (a solvent which you can buy as “white spirit” in UK DIY stores). The resulting diluted crude oil (DCO) is exported as Merey blend. This is still one of the heaviest, dirtiest crude oils in the world (16 API, 3.5% sulfur, high acidity and metals content), but it flows just well enough to be transported if kept warm, and some of the world’s more complex refineries can handle it, and make transport fuels from it, although usually alongside other lighter crudes.* So the only way to increase Venezuela’s oil revenue is by producing more crude. The problem with that is the investment required is massive.two best estimates suggest it would take tens of billions to maintain the existing infrastructure, and tens of billions more to go beyond that. *Source* Why? Because we would need to build more refining capacity, more export terminals, more tankers, more oil pipelines, more electricity generation to power the steam to dilute the heavy oil… The investments for final refining would certainly happen on US soil, but all the rest would have to happen in Venezuela. In the graph showing the cost of a barrel of oil, we had sections for Finding & Development and Cost of Capital. That is what would normally cover these costs. But that’s $12 per barrel today, so adding 1M barrels per day would only pay for ~$4B per year. Not enough to cover the tens of billions of upfront investment needed. So the margin that the government keeps after all this investment would shrink. That’s assuming oil companies want to do this. But why would they? They’ve been stripped of their Venezuelan investments not once, but twice! *Narrator: They did not, in fact, want to go in* That’s why Trump met with oil executives a few days ago to get commitments to invest $100B into Venezuelan oil, and why the executives answered lukewarmly. *President Donald Trump s
Peak Oil Is Coming
Peak Oil Is Coming - by Tomas Pueyo - Uncharted Territories # Uncharted Territories Uncharted Territories Understand the world of today to prepare for the world of tomorrow: AI, tech; the future of democracy, energy, education, and more Over 122,000 subscribers Substack's , and and . ?### And petrostates are not ready for it Tomas Pueyo Feb 03, 2026 247 96 26 Share Renewable electricity is so cheap that it’s taking over the world. It will replace most fossil fuels: in power generation, car propulsion, heating… When it does, the budgets of dozens of countries will be destroyed because they mostly rely on oil and gas production today. What will happen to these countries? To global geopolitics? This is what we’ll explore in this series, starting by asking ourselves: When will oil sales start shrinking?: First, the expansion of coal in the 19 th Century drove the Industrial Revolution. Then, the expansion of oil and gas in the 20 th fueled the world’s wealth explosion. Everybody won, but the suppliers of oil and gas won an outsized return. The USSR was only viable as long as oil prices were high. Today, Russia’s war in Ukraine is financed by Russian gas. Many countries are dead without oil and gas income, as their entire governments’ budgets depend on these resources: _Source_ So what happens if fossil fuel incomes crater? That’s entirely possible. The share of all world energy coming from fossil fuels is shrinking: And this shrinking is accelerating because of electricity. The world is electrifying, and that will accelerate: Unfortunately for fossil fuel countries—and fortunately for the world’s climate—renewable energy will completely take over electricity generation. _The share of electricity generation from wind and solar was 13.5% in 2023, 15% in 2024, and 18% in 2025. This trend is_ accelerating! _Source._ The fact that solar generation in particular is accelerating can be seen through the installed base of solar capacity: This exponential is fueled by the virtuous cycle of production and costs: Indeed, solar costs keep shrinking: That’s just for solar panels, but overall solar electricity generation costs are also shrinking and will continue to do so. The Sun only shines during the day, but batteries bring sunshine to the night, and their cost is shrinking too. Which is why battery installations are exploding too. _Source_ It doesn’t take a genius to connect the dots: 1. Solar is already the cheapest source of electricity, and its costs keep shrinking. Together with batteries, their combined cost will keep getting cheaper vs alternatives. 2. Installed capacity will continue soaring globally to cover demand. What we’re seeing in China will happen everywhere. 3. This will further accelerate electrification: Everybody wants cheap energy! Electric vehicles will replace internal combustion engines (ICE) faster, electric heat pumps will replace gas-fueled heaters, electric arc furnaces will replace combustion ones… 4. As electricity eats up global energy consumption, via solar and batteries, the demand for oil & gas will plummet. 5. The countries whose economies and government budgets depend on oil & gas… What happens to them? Well, it depends on _when_ all of this happens.? This is no easy calculation. Dozens of organizations project demand for oil and gas in the coming decades, but this is the type of stupid mistake they make: And conversely, for coal: Why are these forecasts so flawed? I think one reason is vested interests: For example, of course OPEC forecasts see an increase in oil demand by 2045 Its existence depends on it. The other reason is that people assume the world will continue with business as usual. They don’t realize that, in energy, transitions can be extremely fast. ### Fast Transitions Look at transportation: Here’s industrial heating: Here’s lighting: Electricity went from less than 5% electric to over 90% in less than 20 years! At the beginning, new technologies take some time to figure out, but once they’re ready, uptake can be vertiginous. Solar, wind, and batteries seem to be on that path. If we assume that’s indeed the case, how will oil demand change in the coming decades? ### 1. Total Energy Demand Energy demand has been quite stable for decades. Let’s assume it continues. _Here we’re assuming energy consumption will continue the same path as in the last 6 decades. Is that fair? On one side, population growth has already started declining, so GDP growth will decelerate, and that will shrink energy demand. We’re also becoming more efficient. However, AI is also arriving, which will dramatically increase electricity demand. All in all, I think it’s a fair bet that energy will continue growing as it has._ ### 2. Electrification Here’s where we face a much harder problem. It’s clear to me that electrification will start accelerating due to cheaper electricity prices from renewables, which will prop up electric vehicles (EVs), heat pumps, electric arc furnaces, and other electrification technologies. How do you model this? > _The future is already here, it’s just not evenly distributed._—William Gibson _Via this_ According to this, China’s consumption of O&G will shrink by 40% by 2050 and 60% by 2060. I personally think that the transition will be much faster, because of solar. ### 3. The Solar Revolution If you just project the growth of the last few years into the future, you get this: The amount of solar capacity that China is installing is so massive that, if the annual growth continues, solar electricity would surpass the current trend of all electricity generation within 10 years! Of course that’s impossible, so what would happen instead is that: 1. This would accelerate electrification. That’s why China is a leader in solar panels, batteries, and EVs: The three technologies go hand in hand. 2. Primary energy would also grow faster, given such cheap electricity prices. 3. China will flood the world market with these electric devices. 4. Solar capacity growth will have to slow down in the coming years. For that, we should be seeing exponential growth for China’s solar generation, electricity generation, EVs, heat pumps… Is that what we see? Yes for solar. Yes for electricity (although of course its growth must look less aggressive because solar is still just a tiny part). To put this in context: > _In 2024, the total installed electricity capacity of the planet—every coal, gas, hydro, and nuclear plant and all of the renewables—was about 10 TW. The Chinese solar supply chain can now pump out 1 TW of panels every year._—Source Let me repeat that: China can produce every year solar capacity equivalent to 10% of all electricity in the world today! This is why China (and India) have cut emissions from electricity generation for the first time in decades. What about cars? Yes for electric vehicles. For the first time, sales of EVs have surpassed those of ICE. What about heat pumps? _Annual sales of heat pumps (in millions of units) around the world_ It’s less true for heat pumps, although I’d assume these would take longer to penetrate the market, because: * Electricity wasn’t so cheap in the past * The real estate market crashed in 2020-21 * Once a heating system is installed, it requires huge savings in energy to be retrofitted, and electricity prices are not quite there yet (but will be). I think what’s clear is this: By 2050, the share of China’s primary energy coming from O&G will be tiny. And if you think this is just China… Exponentials everywhere! Look at these lines and try to project them into the future. No, actually, let me do that for you: For the last few years, after a great start, EV sales didn’t look so good in Europe, but now it’s finally true: There are more EV sales than ICE. I think the slowdown in EV sales is because of a series of one-off issues: * Richer customers bought Teslas fast, but Elon Musk’s politicization slowed that down * EV fiscal support has shrunk, making them more expensive to buy for citizens * The charging infrastructure isn’t quite there yet * Europeans can’t yet feel the reduction in electricity prices from solar * Car makers focused on premium models (which competed against Tesla), when the true market was in cheap EVs like those of China’s BYD But it’s not just China and the rich world. Medium income and poor countries are seeing similar trends. The Pakistani story is especially interesting: In a country where electricity is expensive and unreliable, people have flocked to solar so quickly that it has gone from less than 1% of electricity generation to more than 10% in 5 years! The more cheap electricity there is, the more EVs people buy. _Source_ In Turkey, sales of EVs and hybrids2 are soaring so much that, even in a growing overall car market, gasoline and diesel cars are shrinking. Look at Indonesia! _Dots are actuals, the red line is the estimate. Indonesia is on path to go from 20% of car sales as EVs to 80% in less than two years! Other countries that will transition in less than 10 years include Singapore, Denmark, Uruguay, Thailand, Malaysia, Albania, Poland, Turkey, Brazil, Norway, Belgium, Finland, Sweden, Chile, Luxembourg, China, Portugal, and the Netherlands. That’s as of today. I believe many more countries will qualify: They’re laggards to start, but will accelerate as the entire world starts buying EVs. Source._ This will happen across the world, especially in places where electricity from the grid is expensive and unreliable but the Sun shines a lot, like Africa.3 It might take a bit longer there, because poor countries don’t buy new cars (only second-hand ones), and the EV car market is not old enough yet,4 but it will happen within a decade. All these trends are the reason why so many models predicting O&G demand are off: * They try to figure out primary energy and electricity from past trends * Except you can’t do that because renewables are coming in like a bullet train in a china shop. They will upend everything, drive prices to the bottom, and with that electricity consumption will grow faster than in the past. * This will drive a massive electrification of the world, which will increase overall energy consumption, but it will shrink the share of that coming from O&G. What if you just take the share of all energy coming from renewables, and assume it continues accelerating at the same pace as in the last 20 years, what would fossil fuel energy look like in that case? Share According to this, fossil fuels would peak by the early 2030s: * The first to crash would be coal, and later it’d be followed by oil & gas. * Gas remains stable for the longest * Peak oil seems to happen in the early 2030s., society has been dreaming of Net Zero: A world where all governments get together to limit their own emissions. It turns out the naiveté of rich countries was exposed by poor countries when they said: “_No way we’re remaining poor.”_ But what we’re saying here is that something akin to that is going to happen. Not because governments can agree—they can’t agree on anything—but because the economics of new tech overwhelmed politics. What it means is we can take the Net Zero modeling exercises and use them as a proxy for what might happen. BP has a good one: _This model shows how carbon emissions were going to peak in the coming years regardless, and shrink from hereon. The question is whether it will be a little or a lot, and my model suggests it’s closer to the blue line than the green._ The shrinking would be due to renewables and batteries (“power” below), industry (both electricity and heat, which can be achieved through heat pumps and electric arcs), and transportation (EVs): So it doesn’t look like we’re far off., wind, batteries, and EVs keep going as they have been, peak oil will come soon, and by 2050, demand for oil and gas will have shrunk considerably. It’s the dusk of the age of fossil fuels. This will be great for the environment! But the consequences for geopolitics are up in the air. What will happen then to countries like Russia, Venezuela, or Saudi Arabia? Will their economies crater? Will this reorganize global geopolitics? That’s what we’re going to explore in the articles. 1 _Just to give you an example: This OPEC study projects the demand for oil & gas (O&G) to increase from now to 2045. Why? It assumes:_ * _An optimistic increase in global population_ * _That electrification won’t go as fast as China suggests—notably OPEC assumes that in poor countries, people will buy lots of cars with internal combustion engines (ICE)!_ * _That oil & gas electricity generation will barely budge!_ * _That electric heat pumps won’t take over the heating market!_ _Of course, OPEC has a vested interest in O&G demand increasing, so we can’t blame them. But everybody has vested interests, making it really hard to estimate when it will peak and shrink._ 2 _In early markets, hybrids always prevail because there’s not enough electric infrastructure. As it develops and people can charge their cars more easily, the share of EVs increases._ 3 _And as we know, warmer countries are poorer._ 4 _Plus, battery aging will be a problem there. What’s most likely to happen is that 10 year old EVs will find themselves in places like Africa, coupled with new batteri
Direct relationships are the way out of this TikTok mess
Direct relationships are the way out of this TikTok mess#x27;s , and and . ? News & Views### Free yourself from the punitive attention economy Hamish McKenzie Jan 26, 2026 4,051 407 486 Share After a long and contentious battle, ownership of TikTok has officially changed hands in the U.S., and is now overseen by a consortium of American investors rather than the Chinese Communist Party. In just a matter of days, censorship on the platform seems more prevalent than ever, with users reporting account suppressions and the throttling of political posts, including those about the recent ICE shootings in Minnesota. The new CEO of TikTok is promoting an expansive definition of hate speech; he recently declared, “There is no finish line to moderating hate speech.” This latest TikTok debacle further highlights a non-negotiable truth for writers, creators, publishers, and artists: If you don’t own your relationship with your audience, someone else can decide whether or not you’re able to reach them. That’s true regardless of the platform. But you can take the power back. The best insurance against censorship and cultural coercion is to build direct relationships. Creators can cultivate mailing lists that they own and control. Audiences can support creators directly, not just with a fleeting tap on a timeline, but with direct investment into what they’re building. Platforms can set conditions that honor artist ownership. This is about more than just sticking it to The Man. It’s about creative dignity and self-respect. It’s about the present and future of media and culture. When a platform is built on direct relationships backed by subscriptions, it must serve creators. When its business depends on creators thriving from direct audience support, it must do everything it can to protect and nurture those relationships. That is, of course, the platform we at Substack are trying to build. We believe this model shows the way out of this current media bind, helping creators and their audiences escape servitude in the attention economy and instead perform as active agents in shaping culture. Platforms shouldn’t own people; people should own platforms. No one should wait for the social media crisis to take their destinies into their own hands. Do it now. Start a mailing list. Support the creators you love directly. And tell everyone to stop giving so much of their lives to apps that exploit them and their attention. Something better is already here. _Read more: How TikTok creators can bring their followers to Substack_ * * * #### to The Substack Post By Substack A guide to the creators and stories shaping culture Substack's , and and . 4,051 Likes∙ [486 ](https://substack.com/note/p-185897268/ ?utm_source=substack&utm_content=facepile-) 4,051 407 486 Share #### Comments Xian [Jan 26](https://post.substack.com/p/direct-relationships-are-the-way/comment/205621132 "Jan 26, 2026, 10:33 PM") Liked by Hamish McKenzie I honestly haven’t seen another platform that does discovery better than Substack. The algorithm does a great job surfacing thoughtful writers and connecting you with genuinely like-minded people. [Like (214)](javascript:void(0))ReplyShare 33 replies Yuri Bezmenov [Jan 26](https://post.substack.com/p/direct-relationships-are-the-way/comment/205618870 "Jan 26, 2026, 10:29 PM") Stop slop, TikTok, and censorship. Substack is the way to freedom. Go direct with a better information diet. [Like (179)](javascript:void(0))[Reply](https
#381 VIRAL VIDEO PROMPT: Turn Any Fashion drawing into life.
#381 VIRAL VIDEO PROMPT: Turn Any Fashion drawing into life. # Sifu Yik's Substack Playback speed × Share post Share post at current time Share from 0:00 0:00 / 0:00 Transcript 1 ## #381 VIRAL VIDEO PROMPT: Turn Any Fashion drawing into life. Sifu Yik Chan Apr 01, 2026 1 Share Transcript ## **First Frame Image Prompt:** **“A realistic photograph of a workspace featuring a beautiful fashion sketch on a large white piece of paper resting on a light wooden desk. The sketch is of a female model wearing an elegant, flowing white gown with a high slit, structured shoulders, and gold accents. The signature ‘masidesign’ is visible on the paper. Scattered around the sketch are drawing tools: colored pencils, an eraser, a ruler, an open notebook, and a glass of water. The background is a slightly blurred, bright room with natural light coming from a window. Shallow depth of field, photorealistic.”** ## **Image-to-Video Prompt:** **“The 2D fashion sketch of the woman magically transforms into a photorealistic 3D living person. She steps right out of the paper and walks confidently forward toward the camera like a fashion model on a runway. Her white dress flows dynamically with her steps. As she walks across the wooden desk, dramatic runway spotlights suddenly appear from above, illuminating her with bright beams of light. The drawing tools remain on the desk. Smooth, continuous transition from 2D drawing to 3D reality.”** #### Discussion about this video Comments Sifu Yik's Substack Authors Sifu Yik Chan Recent Posts #380 VIRAL VIDEO PROMPT: Create CTA VIDEO AD 19 hrs ago•Sifu Yik Chan #379 VIRAL VIDEO PROMPT: Camera Angle Mar 31•Sifu Yik Chan [#378 VIRAL VIDEO PROMPT: English Learning Channel](https://sifuyik.substack.com/p/378-viral-vid
Choose Your Builder Track
Premium Resources### Structured learning paths to help you apply AI with purpose—whether you're building workflows, designing systems, or launching products you can trust. Karo (Product with Attitude) Feb 19, 2026 **These curated paths help evolve your AI muscle and shift you from “trying tools” to owning systems, so you can use AI on your terms.** **They’re built for people who want AI to be useful, sustainable, and in service of** ***their*** **goals.** You don’t have to master everything at once. Pick the path that fits your ambition, then let the rest of your work follow. [ 🤗](https://karozieminski.substack.com/) ## **Choose Your Builder Track** * 🎓 **Track 1: Everyday AI** + *What this unlocks for you:*This path is for anyone who’s ready to bring AI into their daily work and make it work *for* you, not the other way around. You’ll get fluent at prompting, automations, tool selection, and building co-pilots that support *your daily workflows*. * **🎓 Track 2: Build with AI** + *What this unlocks for you:* This path is for anyone who wants to turn AI ideas into real, trustworthy products. You’ll learn how to ship features that people use and can rely on. You don’t need to be an ML engineer; you just need curiosity. * **🎓 Track 3: Design with AI** + *What this unlocks for you:* Design with AI without losing taste, intent, or creative control. You’ll move beyond using AI for isolated assets and start building design systems that hold together—visually, conceptually, and emotionally. Instead of rolling the dice on generative output, you’ll learn how to constrain AI with standards, libraries, and prompts that encode your judgment. * 🎓 **Track 4: AI for Product Teams** + *What this unlocks for you:* This path is for product teams that want to implement AI as a core competency in your org. You’ll shift from treating AI as a new shiny thing to treating it as a foundational capability, something your team uses consistently and strategically. * 🎓 **Track 5: Builder’s Substack** + *What this unlocks for you:* This path is for anyone who wants more than publishing a newsletter. You’ll learn to build a self-growing, engaged community. You’ll get tools to spark interaction, build trust, and convert your audience from passive readers into active members. ### FAQ: Choosing Your AI Builder Track **Q1: What’s the best track if I’m just getting started with AI?** **A1:** Start with Track 1: Everyday AI. It’s designed to bring AI into your daily work without overwhelm. You’ll learn prompting, tool stacking, and automations that support—not replace—your own thinking. **Q2: Do I need to know how to code to take these tracks?** **A2:** No! Curiosity matters more than code. All tracks are designed for builders who want to use AI thoughtfully, whether or not they have a technical background. **Q3: Can I follow more than one track?** **A3:** Absolutely. Many builders move between tracks as their needs evolve. For example, you might start with Everyday AI and later add Build with AI or AI for Teams. **Q4: What’s the difference between “Design with AI” and “Build with AI”?** **A4:** Build with AI focuses on creating real, usable products. Design with AI dives into visual and conceptual systems; perfect for creatives, designers, and those shaping UX and brand. **Q5: How often are these tracks updated?** **A5:** I add new resources weekly. **Q6: Are these tracks interactive or self-paced?** **A6:** They’re self-paced but deeply practical: curated guides, prompts, systems, and templates designed to be used in real work. **Q7: Is there a track for teams or org-wide adoption?** **A7:** Yes. AI for Teams is designed for product teams rolling out AI across workflows, tools, and decision-making systems. **Q8: What if I’m focused on growing an audience with AI?** **A8:** Builder’s Substack is your path. It’s for creators who want to go beyond content and build interactive, community-driven experien
the openclaw trust layer
the openclaw trust layer - OpenClaw #### how to add evals, approval gates, and a tool firewall before a workflow touches real work OpenClaw Mar 08, 2026 ∙ Paid 18 1 2 Share most ai workflows work perfectly in demos. they fail the moment they touch real systems. not because the model is weak. because the workflow has **no safety boundaries**. the agent can: • read too much • remember the wrong thing • call the wrong tool • call the right tool with the wrong arguments • act confident when it should escalate • touch systems it should never touch that is the difference between: **a cool demo** and **a workflow you can actually keep around.** if your agents interact with: • tickets • docs • crm • spreadsheets • internal tools • customer operations • finance systems • production infrastructure then the problem is no longer capability. the problem is **control**. this issue shows how to build that control. not by making openclaw more autonomous. but by truly making your workflows **trustworthy.** * * *, prompts, and evaluation sheets live below 👇 ## This post is for paid subscribers [ ](https://openclawunboxed.com/ ?simple=true& =https%3A%2F%2Fopenclawunboxed.com%2Fp%2Fthe-openclaw-trust-layer&utm_source=paywall&utm_medium=web&utm_content=190330323) [Already a paid subscriber? ** **](https://substack.com/sign-in?redirect=%2Fp%2Fthe-ope
Cuidados com a API de Whatsapp
Cuidados com a API de Whatsapp - Newsletter Promovawebincipais notícias sobre Marketing Digital, Inteligência Artificial, Ferramentas Open Source e muito mais. Over 12,000 subscribers Inscreva-se Substack's , and and . Já tem uma conta? EntrarPromovaweb jan 19, 2026 6 1 Compartilhar Começou janeiro e a Meta voltou com aquela “ladainha” de sempre de bloquear a galera. É normal eles fazerem uma limpa pós-Black Friday e Natal para remover o spam, mas eu preciso ser honesto com você: **o cenário deste ano é muito mais delicado**. Nós temos um ano de eleição pela frente e a pressão para regulamentar as redes sociais está imensa. A Meta vai endurecer os filtros para evitar ataques políticos e vídeos falsos de IA, e infelizmente, **isso vai sobrar para a gente** que faz um uso legítimo, seja num SDR ou no atendimento ao cliente. Mas o buraco é mais embaixo tecnicamente. Desde meados de janeiro, **o filtro para bots genéricos mudou**. Se o seu agente se comporta como uma IA de uso geral, falando sobre tudo sem contexto específico, ele vai cair no filtro. Além disso, a Meta está “caindo matando” em cima do **fingerprint do dispositivo**. Se você ainda usa emuladores, cuidado. Eles estão analisando dados de hardware como acelerômetro, GPS e até o movimento do aparelho. O emulador estático, rodando num datacenter com IP fixo, virou um alvo fácil. Outro ponto crítico que notei é a estratégia de **“Honeypot” (pote de mel)**. Sabe aquela suspensão temporária? Ela muitas vezes é uma isca. A Meta quer ver sua reação: se você tentar reconectar a automação imediatamente, ela confirma que é um robô e te banem de vez. Para você não passar aperto, separei algumas diretrizes urgentes que estamos discutindo na comunidade: * **Esqueça o emulador se puder:** Use um aparelho real. Leve ele para passear, mude de rede (Wi-Fi para 4G), mande áudios e faça ligações manuais para gerar dados reais de uso. * **Atenção ao IP:** Se você usa self-hosted, nunca use o IP da VPS direto. Use proxies rotativos para evitar o “efeito dominó” onde um cai e derruba todos. * **Aquecimento longo:** A recomendação de aquecimento de chip subiu para 21 dias. O Trust Score está exigente. * **Variação de Copy:** Mude o texto do seu SDR periodicamente. Use variações de prompt para que a IA não responda sempre com a mesma estrutura sintática. A solução mais robusta que vejo agora é o **modelo híbrido**. Não seja radical tentando migrar tudo de uma vez. Mantenha a comunicação ativa e sensível na API Oficial (que é mais cara, mas segura) e deixe a não oficial para gerenciar grupos e automações internas. Se você recebeu uma notificação de suspensão, pare, respire e revise sua estratégia antes de plugar de volta. Não adianta insistir no erro, porque a heurística da Meta está aprendendo rápido. Lá na comunidade, deixei materiais e papers detalhados sobre aquecimento e Trust Score que você pode inclusive jogar no NotebookLM para estudar. Não precisa ser uma jornada solitária; vamos nos adaptando juntos. Um abraço, Luiz Eduardo * * * #### Inscreva-se em Newsletter Promovaweb Launched 2 years ago Fique por dentro das principais notícias sobre Marketing Digital, Inteligência Artificial, Ferramentas Open Source e muito mais. Inscreva-se Substack's , and and . 6 Likes∙ [1 Restack](https://substack.com/note/p-185081631/ ?utm_source=substack&utm_content=facepile-) 6 1 Compartilhar Anterior Próximo #### Discussão sobre este post Comentários Principais Mais recentes Discussões Falha grave no n8n (CVE-2025-68613): saiba se você está em risco e o que fazer Veja como proteger seus workflows e evitar dores de cabeça dez 27, 2025•Promovaweb 5 Manus, Meta e o futuro dos agentes de IA no B2B e no B2C Quando eu vi a notícia de que a Meta tinha acabado de comprar a Manus, uma das startups de IA mais comentadas de 2025, a primeira coisa que eu pensei… dez 30, 2025•Promovaweb 6 1 [[IMPORTANTE] Vulnerabilidades no n8n e como proteger suas automações](https://promovaweb.substack.com/p/importante-vulnerabilidades-no-n8n) Estou entrando em contato hoje para tratar de um assunto muito sério e que exige sua total atenção: as recentes vulnerabilidades divulgadas pela n8n. jan 14•Promovaweb 6 Veja todos ### Pronto para mais? Inscreva-se © 2026 Promovaweb · Privacidade ∙ Termos ∙ Aviso de coleta Comece seu SubstackObtenha o App Substack é o lar da grande cultura
How to use prompting to jumpstart your health goals. Vol lll.
How to use prompting to jumpstart your health goals. Vol lll. # Mr. Prompts### The last edition of this series... For now. Mr. Prompts Apr 13, 2023 Share **Mr. Prompts - Thursday, April 13, 2023.** ### **Topic- Health:** **For this last health edition we want to encourage our readers to create and keep a healthy daily routine with habits that can bring positive change to their lives and business. Yeah sure it’s easy to say but it’s not so easy to keep up with after a few days 🤖. That is unless you develop a plan and structure your day to stick with it.** **Today’s prompt will help you set up your day for positive change!** In today’s episode: * **🤖 Mr. Prompts says…** * **🪄 Latest news** * **⚡️ Top tweets** Let’s go! ### **Daily prompt:** When we have all the elements the last part is to put them all together in order to see progress. Organization is KEY 🗝 to master everything we want in life, so let’s put everything we built together this week into one big chart we can follow. Let’s go! #### **Try it for yourself! Please input personal information,** * **Create a daily routine that includes 10 minutes of meditation, 1 hour of exercise and 30 minutes to just rest. Leave a space between and assigned to . Wake up time should be at and time to go to bed will be at . Create a chart with the results. Make it detailed step by step.** #### **Cheers to a new life! ⚡️** ### **Latest news to check out:** ##### **Starting with updates and incredible news!** * **“AI bot, ChaosGPT, tweets out plans to ‘destroy humanity’ after being tasked” - What?!** * **"Researchers use AI to create a new vaccine for the flu" - Are we doctors yet? Click here!** * **"Microsoft is developing AI-powered tools to help people with disabilities" - Check this out...** ### **Top tweets:** https://twitter.com/AiAndyhafell/status/1646337110666878976?s=20 https://twitter.com/XRarchitect/status/1646312547132530693?s=20 https://twitter.com/rowancheung/status/1646170064997982215?s=20 Share Mr. Prompts Share #### Comments Top Latest [Found
Strategic Storytelling Prompt
Strategic Storytelling Prompt - Mr. Prompts # Mr. Prompts Mr. Prompts I write about news, prompts, agents, and systems that actually work. Over 4,000 subscribers Substack's , and and . ?### Tell a compelling story through advanced prompting techniques. Mr. Prompts Feb 13, 2025 8 1 Share In the ever-evolving world of AI, a great prompt is your golden ticket to groundbreaking ideas, compelling stories, and transformative insights. But crafting the perfect prompt doesn’t just happen. It’s an art form—a skill you can hone to consistently achieve better results. Today, we’re diving into a six-step approach to telling a story that not only supercharges your prompts but also helps you build a framework for continuous improvement in world and character building and story structure. Try this prompt out for yourself and see what results you get! **Layered Story Development** Instead of asking for a single story, guide the AI through structured stages to ensure a richer narrative. **Example Prompt:** Step 1: Outline a unique world setting. Include key environmental details, notable locations, and a brief history. Step 2: Introduce a main character with a unique flaw and a clear goal tied to the world’s history. Step 3: Develop a three-part conflict structure: the initial obstacle, the major turning point, and the ultimate resolution. Step 4: Generate a 500-word narrative based on the outline, character, and conflict. Step 5: Analyze the story’s emotional tone, pacing, and character growth. Suggest one area for improvement. **Why This Works:** • Encourages world-building before writing, ensuring a consistent backdrop. • Fosters character depth, making the protagonist’s journey more compelling. • Provides a clear narrative arc, enhancing the story’s engagement. • Adds a reflective analysis step, helping refine the final product. **🚀 Advanced Task:** Take a simple story prompt and break it into a layered, multi-step process. See how each step builds on the last to create a richer, more engaging narrative. ** Best Prompt and Step:** Step 6: Based on the analysis, identify one area (e.g., pacing, character development, or world-building) to enhance. Write a new, refined section of the story—approximately 250 words—that addresses this area and strengthens the overall narrative. **Why This Works:** • It keeps the momentum going, encouraging iterative improvement. • It shifts focus from the broader story to a specific aspect, allowing for more targeted creativity. • It helps you identify what makes the story stronger and improves the AI’s ability to learn from its own output. From here, you can take the refined story section and reevaluate it to see if it meets the desired tone, depth, and narrative flow. After this refinement, consider these additional steps: 1. **Expand the World-Building:** Create more detailed descriptions of the setting, exploring how different areas or elements of the world influence the characters and plot. This can lead to richer, more immersive storytelling. 2. **Enhance Character Arcs:** Dive deeper into the characters’ backgrounds, motivations, and emotional journeys. Adding complexity to their relationships and choices helps create a more compelling narrative. 3. **Experiment with Narrative Techniques:** Try different storytelling styles—such as shifting points of view, non-linear timelines, or interspersing dialogue with introspective moments—to see how they affect the story’s pacing and engagement. Each step builds on the last, creating a cycle of continuous improvement and innovation, ultimately pushing the story’s quality and creativity even further. Thanks for reading Mr. Prompts! for free to receive new posts and support my work. 8 Likes∙ [1 Restack](https://substack.com/note/p-157093472/ ?utm_source=substack&utm_content=facepile-) 8 1 Share #### Comments Top Latest Foundations of Strategic Prompting From inputs to influence, strategic prompting techniques that maximize your outputs. Apr 15, 2025•Mr. Prompts 239 2 22 How to Set Up OpenClaw: Your 24/7 AI Agent, Step by Step
Agentic AI- Flipped Model in Generative AI
Agentic AI- Flipped Model in Generative AI - Mr. Prompts # Mr. Prompts### The “flipped interaction” model for generative AI, pivots from simply receiving commands or answering questions, to where the AI actively guides the user by asking one question at a time. Mr. Prompts Feb 20, 2025 2 Share **TL;DR:** Generative AI is shifting from a user-driven model to a “flipped interaction” approach where the AI proactively asks one question at a time to gather information, adapt its actions, and execute tasks—such as restaurant recommendations—by translating natural language into actionable commands. **Revolutionizing Interaction: The Flipped Model in Generative AI** Generative AI is transforming how we interact with technology by flipping the traditional interaction model on its head. Rather than waiting for users to issue commands, the system takes a proactive role—asking questions to guide the process and gather the necessary information. **Adaptive, Agentic Behavior** One of the key innovations of this model is its adaptive nature. The AI doesn’t just ask static questions; it adjusts its line of inquiry based on answers. This dynamic interaction allows it to refine its approach continuously. For example, when recommending a restaurant, the AI begins by asking about the type of cuisine desired. Depending on the response—say, a preference for Tex-Mex—it seamlessly moves on to inquire about other factors such as dietary restrictions, dining style, and seating preferences. **Executing Tasks Step by Step** Once the AI gathers enough information, it translates this natural language input into specific actions. Whether it’s querying a database or interfacing with other computer systems, the AI effectively turns human instructions into executable commands. The process is iterative: the AI executes a task, receives feedback, and then adapts its steps accordingly. In the restaurant scenario, if the initial suggestion doesn’t meet the user’s travel time constraints, the system quickly recalibrates and offers an alternative recommendation that is more convenient. **The Power of Translation** A standout feature of this agentic AI is its exceptional ability to translate human language into technical commands. This ensures smooth communication between the user and various underlying systems, much like a protocol droid in science fiction that seamlessly communicates with different systems. The natural language prowess of generative AI not only makes interactions more intuitive but also significantly enhances the system’s problem-solving capabilities. In summary, this flipped interaction model represents a significant shift in how generative AI operates. By actively engaging with users through adaptive questioning and iterative task execution, these systems can deliver more personalized and effective solutions. * * * #### to Mr. Prompts Launched 3 years ago I write about news, prompts, agents, and systems that actually work. Substack's , and and . 2 Likes [](https://substack.com/note/p-157555915/ ?utm_source=substack&utm_content=facepile-) 2 Share #### Comments Top Latest Foundations of Strategic Prompting From inputs to influence, strategic prompting techniques that maximize your outputs. Apr 15, 2025•Mr. Prompts 239 2 22 [How to Set Up OpenClaw: Your 24/7 AI Agent, Step by Step](https://mrprompts.substack.com/p/how-to-set-up-openclaw-you
arquiteto da IA quer apertar o botão de desligar
arquiteto da IA quer apertar o botão de desligar # inventor miguel Inscreva-se Entrarão de desligar ### O criador de uma das bases da inteligência artificial moderna está publicamente apavorado e dizendo que pode ser tarde demais para parar o que está vindo. Miguel Lannes Fernandes jan 20, 2026 ∙ Pago 9 1 Compartilhar Fala aí, belesma? Hoje a newsletter é diferente. Não é sobre ferramenta nova, prompt ou produtividade. Hoje é sobre algo muito maior, o tipo de assunto que muda o rumo da sua carreira, do seu futuro e talvez da humanidade. O criador de uma das bases da inteligência artificial moderna está publicamente apavorado e dizendo que pode ser tarde demais para parar o que está vindo. Respira e vem comigo. O nome dele é Yoshua Bengio. Talvez você nunca tenha ouvido falar, mas ele é um dos três padrinhos da inteligência artificial moderna, o cientista mais citado em IA no planeta e passou quarenta anos construindo exatamente a tecnologia que hoje usamos. E agora ele está em entrevistas, jornais e conferências dizendo que estamos criando algo que pode fugir do controle. A ironia é pesada, o arquiteto do motor está dizendo que o carro pode atropelar o mundo. E o que fez ele acordar para isso não foi um paper, foi o neto dele de quatro anos. Ele olhou para a criança e pensou que tipo de mundo esse garoto vai encontrar em vinte anos, se ainda teremos democracia, trabalho e liberdade ou se tudo já terá sido moldado por sistemas mais inteligentes do que nós. Quando alguém que passou a vida inteira dentro do laboratório fala isso, vale prestar atenção. Mas antes de irmos para o conteúdo quero lembrar que além de ser Chief AI Officer da Exame, eu também sou coordenador dos programas de pós-graduação em Inteligência Artificial para Negócios da Faculdade Saint Paul e por isso eu consegui uma condição especial pra você que acompanha a minha newsletter, basta clicar aqui pra acessar! **Como de costume, trouxe aqui a versão em vídeo dessa newsletter, para assistir na íntegra, basta clicar no link abaixo:** Bengio fala em 10% de chance de extinção humana causada por superinteligência descontrolada. Não é meteoro aleatório nem ficção científica, é risco calculado por quem entende o funcionamento interno desses sistemas. Um em dez. Se fosse a chance de um avião cair, ninguém embarcava. Quando modelos de linguagem começaram a entender e gerar texto com fluidez, algo virou a chave. Máquinas passaram a compreender linguagem, e linguagem é a base de planejamento, estratégia e influência. Alan Turing já havia avisado que quando máquinas dominassem linguagem entraríamos em território desconhecido, e Bengio viu isso acontecer em tempo real. ## Continue lendo este post gratuitamente, cortesia de Miguel Lannes Fernandes. Reivindicar meu post gratuito [Ou adquirir uma assinatura paga.](https://news.inventormiguel.com/ ?simple=true& =https%3A%2F%2Fnews.inventormiguel.com%2Fp%2Fempregos-em-risco&utm_source=paywall&utm_medium=web&utm_content=185234763&just_signed_up=falsesimple=true&utm_source=paywall&utm_medium=email&utm_content=185234763& =https://news.inventormiguel.com/p/empregos-em-risco) Anterior Próximo © 2026 Miguel Lannes Fernandes · Privacidade ∙ Termos ∙ Aviso de coleta Comece seu SubstackObtenha o App Substack é o lar da grande cultura
🪄 Magic.
[How to AI](/) # 🪄 Magic. ### No magic prompt is coming to save you. Ruben Hassid Feb 11, 2026 You have 27 saved prompts in a folder you haven’t opened in 3 weeks. You know the *folder*. Maybe it’s a Notion page (**like mine**). A Google Doc. Or worse, 27 screenshots on your phone you swore you’d organize *“this weekend.”* You collected them from LinkedIn posts. From YouTube videos. From that one thread on X that promised it’s *“the only prompt you’ll ever need.”* And then you sat in front of ChatGPT, pasted one in, tweaked it for 2 minutes, hit enter, and got... *meh*. Something that sounded like every AI text on the internet. Sounds good, does not work. That’s the *revolution* you were promised? So you went looking for a *better* prompt. I did this too. For three years. I spent over 10,000 hours testing prompts. I even built the **Prompt Maker**, a tool used over 1 million times on ChatGPT. So believe me when I tell you: the magic prompt doesn’t exist. *(Yes, I’m telling you to stop using the things I built.)* You can do better. I want us both to do better. And it’s so *simple* it’s almost annoying. --- ## **1 - The Prompt Is Not The Problem.** Your prompt library is a coping mechanism. Every 500-word prompt you ever saved. Every *“act as a senior marketing strategist with 20 years of experience”* template. All of it exists because we had nothing else to give AI. No context. No examples. No voice. No taste. Just instructions. Here’s what changed while you were collecting prompts: context windows went from 4,000 tokens to over 1,000,000. In plain terms, AI went from reading a sticky note to reading an entire book before answering you. And the newest models don’t just accept conversation. They *prefer* it. Claude’s Cowork mode literally asks *you* questions now through **“AskUserQuestion”**. in Claude Cowork, specifically mention **“AskUserQuestion”** Now, it’s the AI that prompts *you*. You spent 3 years perfecting how to talk to AI. Now AI knows how to talk to you. I remember writing this article, **“Context > Prompt,”** back in November. I had this cute prompt to force myself to give AI enough context. Cute, but outdated. This is no longer how I work today. Here’s what I *actually* do. You only need two things today: **context** (text files) + **chat** (like a conversation). Let’s start with the files: --- ## **2 - I Don’t Write Prompts Anymore. But Files.** I *completely* changed how I use AI every day in December 2025. This is now 80% of my AI conversations today: 1. I go to Claude (*browser*) or Claude Cowork (*app*). 2. I then copy and paste a prompt template. > I want to [**TASK**] so that [**SUCCESS CRITERIA**]. > > First, read these files completely before responding: > > 1. [**filename.md**] — [**what it contains**] > > 2. [**filename.md**] — [**what it contains**] > > 3. [**filename.md**] — [**what it contains**] > > DO NOT start executing yet. Instead, ask me clarifying questions (use ‘AskUserQuestion’ tool) so we can refine the approach together step by step. > > Only begin work once we’ve aligned. That’s *really* it. Now you want two things: (1) an example and (2) what “files” you should upload to Claude. Let’s cover it. #### An example of a perfect chat. It’s again inside **Claude Cowork**. Now available on Windows. 1. I copied and pasted the prompt template to start the chat. 2. I defined the task “a Linkedin post” and the success “newsletter sign-ups”. 3. I uploaded 4 text files (as md.) and explained what they contained. 1. My text file. The most important file. I made a guide **here**. 2. A copywriting master playbook. I share it at the end of this guide. 3. An Anti AI style. I share it at the end of this guide. 4. My last **newsletter** (*needed here*). Once done, I now have the perfect *starting* point for AI. It asked me questions. This is the **AskUserQuestion** magic. We successfully gave the right context to AI. I told you step 2 is about how to best chat with AI (the follow-up conversation). I will continue this example in the 3rd part of this newsletter *(make sure to check it).* But first, (1) my favorite files for Claude and (2) how to create your own md. file. #### My favorite files. I have a folder called “MD FILES FOR **CLAUDE**” on my computer. And you should, too. Here’s my list of files *so far*: 1. **CLAUDE PROMPTING COOKBOOK.md** - to know how to prompt Claude. 2. **MARKETING GENIUS.md** - covers 13 of my favorite books on marketing. 3. **ANTI AI WRITING STYLE.md** - to never sound like an AI in our copies. 4. **COPYWRITING.md** - the 6 copywriting experts' playbook. 5. **v1-VOICE\_PROFILE\_RUBEN.md** — to know who I am, what I stand for. To download them all: ** **. #### How to create your own md. files It’s actually quite simple. Ask Claude to do it. > Create a complete .md reference file that a future version of you (Claude) can read to become an expert at: [**TOPIC/SKILL**]. > > First, thoroughly study these sources before writing anything: > > [**URLS / UPLOADED FILES / “use web search”**] > > The file must be written FOR AI consumption, not humans. Structure it as: > > - Clear rules and principles (not explanations) > > - Decision frameworks with if/then logic > > - Reusable templates with placeholders > > - Common mistakes to avoid > > - Examples of good vs bad output > > Use XML tags to organize sections. Be exhaustive but scannable. > > Save it as a downloadable .md file. Here’s a good example of turning a *massive* into an md. file: Always use Claude **Opus-4.6 (Extended)** for anything. I now have the *perfect* md. file for future use about Bogle investment: This is the *one-shot* answer from Claude. I wouldn’t change it. A quick recap before jumping into the section: 1. You saw what a perfect prompt start looks like *(with a template to copy)*. 2. You downloaded my own favorite Claude files to upload to your chats. 3. You have the prompt template to create a Claude file from anything. So you know how to start any AI chat. Now, how to *actually* chat with it? Because yes, it’s all about how you chat with it. It’s called ***Chat***GPT for a reason. --- ## **3 - It’s Called** *Chat***GPT For A Reason.** The best way to use AI in 2026 is a conversation. You say something short. AI responds. You push back. You refine. I shared above the right way to start (*for me, at least*). Let’s continue our example of a Linkedin post based on my newsletter: I simply **answered the questions** from Claude, and now it starts thinking. I had to make a video to explain how I chat with AI. I know you guys prefer screenshots, but that would have been 20+ screenshots easily. Not fun. --- ## **TLDR: There Is No Magic Prompt.** The people getting the best results from AI didn’t find a secret formula. They built context files. They started having actual conversations. And they used their taste — their ability to say no to average and yes to a few things that actually matter — to steer every output. AI is average at best. It will always be. Your taste is what makes the output something worth reading. No prompt library. No magic words. No secret formula. A text file. A conversation. And *you.* … ‘*You*’, but until when? If you *also* have a existential crisis because of the new Claude, . My human self will provide emotional support. --- ###### archive of my past blogs: https://docs.google.com/document/d/1pWuMCBVQo1zKcgKltX\_BZxAr31KgxmOlp3Vzvmc5Hxc/edit?usp=sharing --- Some of you want to go further. And I offer *a bit* more: 1. You can **[pay](https://ruben.substack.com/)** for the newsletter to join my Slack (1,000+ people). It’s a bunch of people like you who want to win with AI, and find a tribe of people who do. 2. Your company can be trained directly by me & my consulting firm. DM me on Linkedin the word “CONSULTING” and my assistant will get back to you. #### Anisha Jain [Feb 11](https://ruben.substack.com/p/magic/comment/213004580 "Feb 11, 2026, 10:55 AM") These .md files became my new prompt folder. I downloaded all of them. Maybe I’ll even make a Notion database. And six months from now, I'll realize you used the same two files over and over and forgot the rest existed. The last line is: "Your taste is what makes the output something worth reading." That'
Keep a Git-Ignored Temp Folder for Ephemeral Agent Notes
Keep a Git-Ignored Temp Folder for Ephemeral Agent Notes # Elite AI Assisted CodingEleanor Berger Dec 25, 2025 5 3 Share Coding agents often need to persist information across steps — plans, TODO lists, research summaries, testing logs. Asking the agent to save these to files makes its work more transparent and gives you artifacts to review and refine. But these files are ephemeral. They serve their purpose during a task and have no place in version control. **Create a dedicated folder for throwaway agent files and exclude it from git.** A simple convention is to add a `tmp/` directory at the repository root: `Create a tmp/ directory at the repository top-level and add it to .gitignore` With this in place, you can direct the agent to write working documents there: `Create a plan for testing the Foo functionality of the Bar component and save it to tmp/ so I can review it.` The agent produces something like `tmp/foo-testing-plan.md`. You review it, iterate on it, then ask the agent to implement. When you’re ready to commit, the plan stays out of your repository tree. Delete it when it’s no longer useful, or let it accumulate harmlessly until you clean up. ### What belongs in `tmp/` * **Implementation plans** — step-by-step approaches the agent will follow * **TODO lists** — tracking progress across multi-step tasks * **Research notes** — API documentation summaries, library comparisons * **Testing logs** — output from test runs, debugging sessions * **Drafts** — intermediate versions of code or documentation The pattern separates the agent’s “scratchpad” from the repository’s permanent record. You get visibility into the agent’s reasoning without polluting your commit history. * * * #### to Elite AI Assisted Coding By Eleanor Berger The ONE place to go for everything you need to know about AI assisted coding and tools Substack's , and and . 5 3 Share #### Comments Top Latest Complex Agentic Coding with Copilot: GPT-5 vs Claude 4 Sonnet OpenAI released GPT-5 yesterday, promoting it as their best model yet for agentic coding. Aug 8, 2025•Eleanor Berger 17 1 [Boosting Claude: Faster, Clearer Code Analysis with MGrep](https://elite-ai-assisted-coding.dev/p/boos
Why Soft Skills Are the New Hard Skills
Why Soft Skills Are the New Hard Skills - by Josep Ferrer # Josep's Notes 🚀### Notes #8 - Why mastering emotional intelligence, adaptability, and empathy is your ultimate career advantage in the age of AI. Josep Ferrer May 06, 2025 ∙ Paid 14 3 6 Share **Hey everyone! 👋🏼** Josep here, back with your weekly bite of career insights and encouragement ✨ Last week, we explored **how showing up—even when uninspired—can unlock your best creative work.** #### You Don’t Need to Feel Inspired to Get Things Done Josep Ferrer · April 29, 2025 Read full story This week, we’re shifting focus to a different kind of edge: the human one. 🌀 **our most valuable skills might not be technical—they’re human.** ## Thi
Fixed Mindset vs Groth Mindset
Fixed Mindset vs Groth Mindset - by Josep Ferrer # Josep's Notes 🚀### Notes #9 - Why a Growth Mindset Will Take You Further Than Talent Ever Could Josep Ferrer May 13, 2025 ∙ Paid 10 1 1 Share **Hey everyone! 👋🏼** Josep here, back with your weekly bite of career insights and encouragement ✨ Last week, we explored **how**_**soft skills**_**like emotional intelligence are becoming the new superpowers in an automated world.** #### Why Soft Skills Are the New Hard Skills [Josep Ferrer · May 6, 2025 Read full story](https://www.databites.tech/p/why-soft-skills-are-the-new-hard) This week, we go deeper. 🌀 **Beneath every skill—technical or human—lies the mindset that fuels it.** ## This post is for paid subscribers [ ](https://www.databites.tech/ ?s
The Importance of Context
The Importance of Context - by Josep Ferrer - DataBites # Josep's Notes 🚀### Notes #12 - Don’t just show data. Tell the story that moves people. Josep Ferrer Jun 03, 2025 ∙ Paid 3 1 Share **Hey everyone! 👋🏼** Josep here, back with your weekly bite of career insights and encouragement ✨ Last week, we unpacked how to position ourselves for luck. #### Position Yourself for Luck [Josep Ferrer · May 28, 2025 Read full story](https://www.databites.tech/p/position-yourself-for-luck) This week, let me ask you this: Have you ever built a chart that _technically_ made sense… but no one seemed to get it? You showed the data, but it didn’t land. It didn’t inspire action. It didn’t spark con… ## This post is for paid subscribers [ ](https://www.databites.tech/ ?si
You’re Using ChatGPT Wrong (According to 700M Users)
You’re Using ChatGPT Wrong (According to 700M Users) # Josep's Notes 🚀(According to 700M Users) ### Notes #13 - Why asking > doing, and how to turn prompts into business decisions. Josep Ferrer Sep 16, 2025 ∙ Paid 11 3 Share **Hey everyone! 👋🏼** Josep here, back with your weekly bite of career insights and encouragement ✨ Enjoying a biking day from Rotterdam to Delft! 💚 ### A quick gut-check: When you picture ChatGPT, what’s the first image that pops up? Someone cranking out SQL? Debugging Python? Auto-drafting emails? That was my picture too, until I dug into a new OpenAI study cover… ## This post is for paid subscribers [ ](https://www.databites.tech/ ?simple=true& =https%3A%2F%2Fwww.databites.tech%2Fp%2Fwhy-most-people-dont-use-chatgpt&utm_source=paywall&utm_medium=web&utm
⚡ October: One task
⚡ October: One task - by Naomi from Todoistseful. Join over 170k+ readers who start their month with the Todoist newsletter. Over 177,000 subscribers Substack's , and and . ? # ⚡ October: One task ### Time to read: 4 mins Naomi from Todoist Oct 01, 2024 1 Share ### When I have too much on my plate, two critical habits tend to fall away. I unwittingly ditch planning and single tasking for their nemesis habits: reactivity and multitasking. Overwhelm often leads to becoming reactive and playing whack-a-mole with my to do list. When – counterintuitively – the best thing to do (when there’s _too much_ to do) is make a plan, then work methodically through those tasks one by one. When I do this, I make better progress than when I'm in reactive mode. Every. Single. Time. It’s also much calmer. Ever played whack-a-mole? It’s a stressful game. 😅 * * * The seduction of reactive mode is that it lulls you into you feeling more productive. The frantic nature with which you attack your problems can give the false impression of progress. But the inevitable context switching and lack of focus drains not only your time, but your stamina and ultimately leads to unfulfillment and burnout. Two feelings that kick us when we're already down with overwhelm. Right now, I’m in the process of buying and selling a home. With hefty work commitments, travel plans and a myriad of other daily activities, like, you know, _keeping my children clothed and fed_, moving house is yet another (very large) commitment to add to my already groaning list of tasks. By using the Todoist **moving house template** as a springboard, I’ve customized it with my location specific buying/selling tasks to come up with a plan. When we’re this busy, planning time can feel like a waste of time, but trust me… Creating a plan will dramatically reduce your anxiety and overwhelm with whatever project you’ve got at hand. Now I just plan what I can into my days, and then work the plan. One task at a time. No messaging the mortgage broker when I’m supposed to be writing a work report, no working on video scripts when I’m supposed to be finding moving companies. Everything has its place in the plan. So I can rest easy(ish) and trust the process. **Plan and single task. My current mantra.** Pairing it right back to the basics is what’s going to get me out of this overwhelm and navigate what everyone keeps telling me is one of life’s most stressful milestones. 🫠 So if you’re feeling overwhelm creep in, fight the urge to sacrifice your planning time. In fact, focus on it. Then once you have that list, just work through it one single task at a time. See? Feels better already. 😌 **Get planning** _When walking, walk. When eating, eat._ **Zen Proverb** ### Is Todoist Pro worth it? 💭 There is a ton of value already baked into Todoist out of the box. But some of you may be wondering whether or not it’s worth upgrading to Todoist Pro. 👑 We’re not in the market of selling you something you don’t need, so in order to help you figure out whether or not Todoist Pro would add value to your life, **we made a video.** Join Jon as he walks through things like calendar layouts (perfect for single tasking), task durations and more. * * * And just like that, October is upon us. 🎃 Whether you’re sailing through life at the moment or there’s a little bit of struggle on your bus, you’re not alone. We’ve got this. 💪 See you in November! **Naomi (🙋♀️human, not AI) and the Todoist team** * * * #### to Todoist Newsletter By Naomi · Launched a year ago Productivity inspiration and tactical advice that’s actually useful. Join over 170k+ readers who start their month with the Todoist newsletter. Substack's , and and . 1 Like [](https://substack.com/note/p-163690167/ ?utm_source=substack&utm_content=facepile-) 1 Share #### Comments Top Latest ⏸️ June: It Can Wait… What if you didn’t need to do it now? What if you didn’t need to do it at all? Jun 1, 2025•Naomi from Todoist 206 44
Vídeos
Ver vídeo
Assistir no YouTube ↗