CTO & Co-founder @ PACO | Building the human-agent collaboration layer | Stanford GSB, MIT TR35
Joined March 2012
- Tweets1.4K
- Following142
- Followers1.4K
- Likes8.5K
My AI agent bought me the Fender Strat I've wanted since I was 15. At 40% of retail. I never asked for it.
I had only said: find me deals on Marketplace, things I might like. It found a cream Strat two blocks away, figured out I wanted it from everything else it knew, negotiated, asked for photos, set the pickup.
Agents doing what we say is table stakes. The magic is when they find what we never said. Waitlist: heypana.ai
Adolfo M.Valdivieso retweeted
I'm starting 𝐆𝐫𝐚𝐝𝐢𝐞𝐧𝐭 𝐀𝐬𝐜𝐞𝐧𝐭 - a hiking series for AI builders in the SF Bay Area.
A lot of my favorite conversations with founders and operators happen over dinners or happy hours but some of the best ones lately have happened on the move. I wanted to add another format to the mix. Every other month (or so), I'll get a small group together to conquer a hike.
🤖 Made with AI
I completed 18 pull requests during Bad Bunny's concert!
How? Well, at Turbo, we built a lot of tools that allow us to work remotely and asynchronously both for our product and our clients directly from Slack. Just like a teammate.
However, you can achieve the same on-the-go productivity by running Claude Code on your computer and this setup: Tailscale creates a secure connection to your computer. Termius gives you a mobile SSH client that doesn't suck. Tmux keeps your sessions alive when signal drops.
Adolfo M.Valdivieso retweeted
Watch Cursor build a 3M+ line browser in a week
We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week.
It's 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM.
It *kind of* works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.
New CodeRabbit report shows AI coding assistants increase bug density by 1.7× compared to human-written code.
For production code, human judgment and testing remain irreplaceable.
Source: lnkd.in/ecSAC7pi
The shift from "write code" to "prove your code works" is becoming critical as Al tools flood repositories with solutions that look correct but may not be.
Coding is not the bottleneck anymore, verification is.
Source: simonwillison.net/2025/Dec/1…
Adolfo M.Valdivieso retweeted
best way to get hard things to work is to first focus on getting all the easy things to work perfectly
Addition is all you need???
BitNet b1.58 introduces a new variant of LLMs with parameters that can only adopt values of {-1, 0, 1} instead of high-precision floats.
BitNet b1.58 not only matches but in some instances outperforms full-precision LLMs.
lnkd.in/e24yGTİK
Diving into the depths of a lesser-known error in modern AI models with Evan Miller's 'Attention is off by one'. #AI #Tech #Innovation 🧠💡 evanmiller.org/attention-is-…
Are you intrigued how GPT-3.5 and GPT-4 performances fluctuate over time? 🧐 Solve math problems or generate code, their abilities significantly vary! Discover why cautious security additions affect these LLMs! ⛑️ Check out the full paper: arxiv.org/pdf/2307.09009.pdf #GPT #ChatGPT
Adolfo M.Valdivieso retweeted
Has anyone ever gotten ChatGPT to say something like: “Hm, that’s a good point. You know what? I’m not sure.”
