Tech design: AI feedback system #8525
Conversation
| DeveloperAgentContext developer_agent_context = 12; | ||
|
|
||
| // NEW: Optional feedback context. If provided, records feedback | ||
| // and returns immediately. Attribution prediction runs async. |
There was a problem hiding this comment.
What does it mean that it runs async? It doesn't support altering tool call results after creation, and doesn't support concurrent completions on one session, so might be safest to run it sync.
There was a problem hiding this comment.
It doesn't support altering tool call results after creation, and doesn't support concurrent completions on one session
Ah, I see that now.
My UX concern was wanting to avoid blocking the user while attribution runs. Another thought I had was that perhaps we should use a "fast" model for the attribution prediction, but I didn't want to add tool-specific model selection as part of this PR. I can run this synchronously in the first pass, and we can consider optimizations as a follow-up.
For review. Not to merge to
main.