Pietro Di Bello

Post-it everywhere @xpeppers #retrospective

An important “context engineering” principle I’m learning these days: 💡Any tool that your AI agents run to check their progress (tests, linting, compilation/build, formatting, running the app) should be quiet on success and succinct on failure

Pair programming with Claude Code Pro be like: instead of a 🍅 Pomodoro timer to switch roles, you just wait until the credit runs out 🤣

7/7 Understanding this architecture makes you a better prompt engineer. You aren't just "talking" to a bot; you are managing its memory and processing power. Next time your LLM struggles, ask: Is this a "Knowledge" problem or a "Working Memory" problem? 🧠🤖

6/7 🚫 Why they both fail When an AI "hallucinates," it’s usually one of three "human" errors: LTM Fail: Didn't know the facts. STM Fail: The info was pushed out of the context window. Working Memory Fail: The logic was too "spaghetti" for the attention mechanism to map.

1/7 🧵 Ever wonder why an LLM "forgets" the start of a chat or struggles with complex code? It’s because AI architecture mirrors human cognition almost perfectly. Let’s break down the parallel between the human brain and Large Language Models. 👇

“Good for humans” code is also “good for AI” code Clear naming, small functions, explicit control flow, and modular design reduce short-term memory (STM) and working-memory pressure for both engineers and models.

Works with Claude Code, Copilot, Cursor & 30+ agents

🚀 Just released: Perplexity Web Research Agent Skill Bring Perplexity's reasoning models to your AI coding assistant: ✅ Current knowledge lookups ✅ Independent second opinions ✅ Deep reasoning delegation Install: npx skills add xpepper/perplexity-agent-skill

aicoding.leaflet.pub/3mbrvhyye4k2e by Chad Fowler, it beautifully correlates XP and Generative AI. Besides the beautiful description of what XP was and why it was a true revolution, it's interesting to view with the same eyes the "generative ai" revolution and warn about what we may miss.

@gicappa Brand: Google Model: Nano Banana 🍌 🤗

@pierodibello Interesting! How does it work?Brand model?

My new keyboard for 2026 😅

I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit. My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently. So, here goes.

This kind of interaction is providing a lot of lessons and learning opportunities, and it allows me to dig deeper with more questions or challenge its suggestions. Here's an example of "digging deeper with more questions":

I also instructed the model to provide one or more tests to validate its proposal (when it makes sense). I then implement the suggestion and update the canvas to get feedback on how I addressed it...

The new canvas mode of @OpenAI 's ChatGPT opens up many new opportunities! For example, I'm using it to practice Rust through small exercises. I prompted it to review my code and suggest refactoring opportunities or functional improvements, without providing a full solution...