Federico Cargnelutti

Should AI-first companies invest more in "AI expert personas" as decision interfaces? For ex: finance, legal, customer, innovator personas. I like this quote: "Skills are tools. Personas are lenses. Agents need both."

Last year I wrote "Turning an LLM prompt into a multi-agent discussion", with multiple AI personas working as a decentralised reasoning engine. Today I saw a great example of this from @takechanman1228 github.com/takechanman122…

I like what @vladquant did with his news app Kagi, he open sourced the UI. Before GenAI, doing this allowed others to contribute to the project and improve it. But today, AI can rewrite the UI completely and adapt it to any use case.

AI doesn't care about building apps, it cares about solving problems. If I want it to show tech news on my reTerminal E1001 dashboard or a Linux TV, it will build the app for that. Here's an example using news.kagi.com/tech/latest. This is a big shift in UX.

Garry Tan's AI toolkit/process, gstack, now has more stars than Ruby on Rails on GitHub. It's a set of skills that turns your computer into a virtual engineering & product team. You can use it with Conductor. github.com/garrytan/gstack

The Arduino Ventuno Q is coming soon. Their AI board is expected to be available by the end of June. You can join the waitlist on the official Arduino Store for a 10% discount: arduino.cc/product-ventun…

The Anthropic Economic Index report: Learning curves anthropic.com/research/econo…

Here's an example. NotebookLM created a conversation between two AI voices explaining my tweets

If devs focus on building small services exposed through APIs instead of full apps, AI + Skills (as .md files) become the interface where intent, logic, and context are defined. We are no longer designing screens, we're designing capabilities. platform.claude.com/docs/en/agents…

This is a great post on how to become AI-first. Also, having a shared repo of skills (.md files that teach Claude specific workflows) improves performance drastically. Knowledge becomes something the whole team can share, use and reuse 👏

👉 An AI agent flow would look something like this: 1. Read the doc and external data 2. Propose changes directly in the doc 3. Trigger an approval request via the Drive API 4. Human reviews, compares via version history, and approves

👉 If you're using Google Docs, you already have what you need: "Suggesting mode" to propose changes, "Version history" to compare them, and "Approvals" to review and approve them before they're applied.

Keeping documentation up to date shouldn't rely on people constantly updating it in real time, it should happen in the background with AI agents continuously reviewing and improving it.

𝗥𝗔𝗚𝗙𝗹𝗼𝘄: ragflow.io It fixes one of the hardest problems in AI, turning messy documents into structured knowledge the model can actually understand.

𝗟𝗹𝗮𝗺𝗮𝗜𝗻𝗱𝗲𝘅: llamaindex.ai It removes the pain of turning messy data into something useful.

Here are some AI projects on GitHub I found interesting and had fun using 👇

The way errors are logged needs to change. Logs are now the notes AI reads to understand problems. The more context, the better. Instead of "Order 1234 not found", log structured errors that explain what failed, why, where, what was expected, and what to do next.

@GergelyOrosz Luckily for those of us using Nest thermostat, it runs on Google Cloud 😂

Some of the most unexpected fallouts of an AWS outage: 1. Postman stopped working: an API development tool built to be an alternative to simple tools like curl. 2. Eight Sleep customers couldn’t use their beds to change positions / set or keep temperature