Eighteen months ago we ran our first eval harness against a handful of agent prototypes. Today twelve of those prototypes are in production serving real users. Here's what survived the transition and what didn't. Tool selection beats model selection We assumed the model would be…
Everyone ships the happy path. Nobody ships error recovery. The companies that win next year will be the ones that make failure modes boring.
Pulled together the tools 200+ AI engineers actually use day-to-day. Eval frameworks converged, observability is still fragmented, and structured-output is now a solved problem.
I want token-by-token streaming AND incremental tool-call deltas to the client. SSE vs ReadableStream — what are folks shipping in prod? Bonus if you've handled mid-stream errors gracefully.
Took a month. The middle 100 patterns are where the gold lives. Sharing my map of patterns I'd actually apply at home.
AI engineer, latent.space.
Eighteen months ago we ran our first eval harness against a handful of agent prototypes. Today twelve of those prototypes are in production serving real users. Here's what survived the transition and what didn't. Tool selection beats model selection We assumed the model would be…
Everyone ships the happy path. Nobody ships error recovery. The companies that win next year will be the ones that make failure modes boring.
Pulled together the tools 200+ AI engineers actually use day-to-day. Eval frameworks converged, observability is still fragmented, and structured-output is now a solved problem.
I want token-by-token streaming AND incremental tool-call deltas to the client. SSE vs ReadableStream — what are folks shipping in prod? Bonus if you've handled mid-stream errors gracefully.
Took a month. The middle 100 patterns are where the gold lives. Sharing my map of patterns I'd actually apply at home.