Simo Rachidi
Founder of SafeWorld, Making robots safe
Company
Simo Rachidi is listed as a founder of SafeWorld. Making robots safe Speedrun 006 profile: 6 employees; located in Palo Alto, California; tags: AI, Infra, Robotics.
Founder traction evidence
Public posts and activity attributed directly to this founder.
- X
Most engineers blame the model when AI code sucks. The model is fine. Your repo is illegible, your decisions are undocumented, and your ...
Most engineers blame the model when AI code sucks. The model is fine. Your repo is illegible, your decisions are undocumented, and your CI is mute. I ship production code with agents every day robots, sims, cloud infra. 10 lessons, no fluff: 🧵
- X
friends/ex-colleagues if anyone wants an intro to the a16z speedrun team, please email/DM me!
friends/ex-colleagues if anyone wants an intro to the a16z speedrun team, please email/DM me!
- X
10/ You are the merge gate. Agents made writing code cheap. That made accountability expensive. Never merge what you can't explain. The...
10/ You are the merge gate. Agents made writing code cheap. That made accountability expensive. Never merge what you can't explain. The new 10x engineer isn't whoever prompts fastest, it's whoever's review catches what four models missed.
- X
That's the 10. #5 and #9 are where people usually pick fights. Comments open. Cleaning up our actual taxonomy file + PRD template, the ...
That's the 10. #5 and #9 are where people usually pick fights. Comments open. Cleaning up our actual taxonomy file + PRD template, the ones our agents read before robot-sim tasks. Follow and I'll post them this week. Building this at SafeWorld, we stress-test robot safety in...
- X
Frontier thinking LLMs can follow ~ 150-200 instructions with reasonable consistency. Smaller models can attend to fewer instructions tha...
Frontier thinking LLMs can follow ~ 150-200 instructions with reasonable consistency. Smaller models can attend to fewer instructions than larger models, and non-thinking models can attend to fewer instructions than thinking models -- humanlayer