Notes

Ideas on product, AI, and building things. Shared first on X and LinkedIn.

In the original language. Each post links to the conversation.

12 of 85 posts

AI still can't beat humans... Congratulations to whoever made that:

Open on X

Making the prototype got cheaper. The hard part moved to framing the problem, reading the feedback and deciding what should change next.

Open on X
▶ Watch video

I don't agree, I would bet is better for you to have your own hardware (if you can afford). What is interesting is that even OpenAI is buying mac minis as I posted days ago: https://x.com/ric_guia/status/2107832567030591892?a=wRkzU2ChescilOY7QKI&s=20…

Open on X

I spent the last long weekend setting up my first autonomous AI agent on a computer dedicated to it. This is not a chatbot. It is not an assistant either.

Open on X

Making the prototype got cheaper. The hard part moved to framing the problem, reading the feedback and deciding what should change next.

Open on X

Does anyone remember the book cracking the code interview? I think this is dead now...

Open on X

In September I launched Radar de Obras: a map of the 130,581 projects registered in Obrasgov. With the research done, the build took about four hours. The first version, which looked finished, had 298 projects.

Open on X

An eval is how you measure whether what the AI produced is good, against a standard you defined beforehand.

Open on X

Two people can pay for the same subscription and work in completely different ways. One uses AI as a search box. The other coordinates tasks that run while they sleep. Keep moving away from just the search box!

Open on X

Hermes tutorials in Portuguese are still rare. So I kept this one complete, so you don't have to stitch together pieces of English videos to make it work. https://inteligenciaabrasileira.substack.com/p/como-montar-o-hermes-agent-do-zero…

Open on X

The most expensive AI in the world running without context will deliver less than an average AI that knows who you are. Build your own second brain!

Open on X
Image

Buying an RTX Spark Surface for local AI? The memory tier decides it. 24GB ($2,599) runs Qwen3.5 9B, 48GB ($3,999) runs Nemotron 3.5 Bolt, 128GB ($5,899) runs Nemotron 3 Super 120B. These are planning estimates, not benchmarks:

Open on X

For ideas that need a little more room.

Read the articles