Dusan Millenkovic
@dusanmilenkovic.bsky.social
The good idea comes true with passionate people.
Code reviews shouldn’t be about style.
They should be about future cost:
– Can this break silently?
– Can we debug it fast?
– Will this scare the next dev?
They should be about future cost:
– Can this break silently?
– Can we debug it fast?
– Will this scare the next dev?
February 11, 2026 at 3:42 PM
Code reviews shouldn’t be about style.
They should be about future cost:
– Can this break silently?
– Can we debug it fast?
– Will this scare the next dev?
They should be about future cost:
– Can this break silently?
– Can we debug it fast?
– Will this scare the next dev?
Most “AI integrations” fail because teams treat LLMs like deterministic APIs.
Prompt versioning, evals, and fallbacks matter more than model choice.
Prompt versioning, evals, and fallbacks matter more than model choice.
February 11, 2026 at 3:39 PM
Most “AI integrations” fail because teams treat LLMs like deterministic APIs.
Prompt versioning, evals, and fallbacks matter more than model choice.
Prompt versioning, evals, and fallbacks matter more than model choice.
Logging every request ≠ observability.
Good observability answers questions you didn’t know you’d ask.
Traces > metrics > logs, in that order.
Good observability answers questions you didn’t know you’d ask.
Traces > metrics > logs, in that order.
February 11, 2026 at 3:39 PM
Logging every request ≠ observability.
Good observability answers questions you didn’t know you’d ask.
Traces > metrics > logs, in that order.
Good observability answers questions you didn’t know you’d ask.
Traces > metrics > logs, in that order.