

Code Club #5: Observability for a production AI agent: what we built and what bit us
Productive's AI agent runs in production. The logs say what it did. They never say why.
We built tooling to answer that, including the parts we got wrong.
It started with a table of traces hardly readable. Then a CLI that ran into rate limits, moved to a shared server so the whole team could use it, only to find that we were missing proactiveness. Then the harder half: going through thousands of sessions, then enlisting help from AI, and working out whether to believe it.
Then what any of it is actually for: traces that turn into bug fixes, and traces that turn into feature requests for the product team.
It ends with one real conversation, walked end to end, and what we changed because of it. For anyone who wants to build the same thing.
🎤 Speaker:
Tibor Rogulja - AI Engineer, Productive
🗓️ September 15th
🕕 18:00
📍 CroAI HQ (Zavrtnica 17)
🔗 Registrations via Luma - https://luma.com/gy583qmf
If you’re building AI agents for production, this one is for you. 👀
#CodeClub #AIAgents #CroAI