felipe
@felipearcela
Followers
69
Following
2K
Media
3
Statuses
137
uc berkeley || tako
San Francisco
Joined July 2019
Puentes applications are open. 10 fellows from our first two cohorts are already living in the US. They've become close friends and work at some of the best AI companies in the world. Apply until April 3rd.
29
55
215
It’s been a few months since Puentes by @AntigravitySF. These are the stories of the fellows and why the program meant so much to them.
14
25
124
if this guy had claude code access we would be living in a different world
95
73
3K
AI agents can't learn from experience. Until now. Earlier this year, one of our customers asked us: "If your web agent uses my website once, will it be easier the next time it visits?" The question highlighted a fundamental gap between human intelligence and AI agents: Humans
arxiv.org
Learning how to do things from trial and error in real time is a hallmark of biological intelligence, yet most LLM-based agents lack mechanisms to acquire procedural knowledge after deployment. We...
36
131
1K
I’m done using ChatGPT to study. Branching sucks. So I built a new interface for thinking.
1
1
3
Practice any art. Not to get money and fame, but to experience becoming. To make your soul grow.
223
9K
41K
We did it. We built Cursor for Google Docs. Introducing TypeOS.
177
75
2K
best moment I've had since I came to SF, thank you @GadiBorovich for the invite and @pedroh96 and team brex for taking the time :)
Day 4: The engineers and @pedroh96, founder & CEO of Brex. If you’re following Puentes and dream of being a part of this room, start getting ready. We’re doing next cohort of Puentes by @AntigravitySF in April 2026.
0
0
2
harvard and its high ceilings always make you think you can do everything
0
0
0
4/ a lot of errors using vector db with prisma and supabase. next time I will not use prisma. our migrate dev broke 4 times, more than 3 hours just fixing migrations. another big challenge was orchestrating the solution itself. we tried to build something too complex instead of
3/ after 2 hours trying to parse, index, and embed a GitHub repo, it’s finally working. mastra workflow helped me debug and structure everything, even without any AI in the loop. next step: merge all branches (we’re building in 4, each with its own DB schema). we’re solving
0
1
1
Rethinking decision-making at the @supabase hackathon Linear + your codebase + Zendesk: finally see the actual complexity and priority of every task on your roadmap Hourly updates 🧵
1
1
10
@ThePrimeagen I also found that insults make LLMs perform much better
0
1
13