Agents that close the loop: lessons from 60% auto-resolution
What it took to move an AI support suite from demo to resolving the majority of tickets on its own.
Read articlePrompts change; behaviour drifts
A one-word prompt tweak can fix one case and break ten. Without a test set, you only find out from users.
Build evals from reality
We collect real inputs, label the expected outcome and score every change against them. A few hundred good examples beat thousands of synthetic ones.
Run them in CI
Evals run on every pull request, just like unit tests. A regression blocks the merge.
Have a question about this, or a project where it applies? Write to us — Dmytro reads every reply.
