
AI engineer @wonder | ex Tesla
Manhattan, NY
Joined January 2017
- Tweets772
- Following827
- Followers1.2K
- Likes52.5K
Had a road trip this weekend with FSD, and Grok was such a game changer in the car. I could describe what I wanted to eat, what kind of detour I wanted, or whether I wanted the scenic route versus the regular route. It was able to handle that flawlessly.
If you’re building AI agents, I would highly recommend Langchain and Langsmith.
🍽️ @wonder is building an agent that plans, orders and ensures delivery of all 21 of your weekly meals, while handling all customizations and customer service resolutions.
With LangSmith, they swapped log debugging for LLM-as-judge scoring. A PM can file a bug and Claude Code fixes it from the trace.
⏯️ @ktech9999’s story
The last three fully new vehicles that @Tesla has launched don’t use a liquid paint shop:
• Tesla Semi (Powder coated)
• Cybercab (Reaction Injection Molding)
• Cybertruck (Bare stainless steel)
After spending lot of $ on grok bot extra usage, last night I wasted sleep trying out Hermes bot and openbot.
1 - Hermes bot was horrible. Even trying to get a bot to make another bot took multiple tries. I had to keep pinging it to check if it’s actually doing any work. The bots would also not know of each other once they’re made you have to tell the main Hermes bot to make them talk to each other. The complexity of setting them up on discord is too high.
2 - open bot didn’t even work.
Spend an hour trying to get it to work with my claude helping me debug. It just would show thinking and not reply.
So
Its too complicated for regular users if someone as technical as me couldn’t get it to run.
As the latest Claude models have shown capability ≠ usability.
This is specifically applicable in user facing apps.
That’s why you should always have a test suite for the model before updating it that’s focused on your use case rather than benchmarks.





