[01] BLOG
NOTES FROM THE RUNTIME.
Engineering write-ups, product decisions and field notes from teams running agents in production.
Engineering
Retry budgets: why one global limit wasn’t enough
A single retry setting looked simple until one flaky upstream started eating whole runs. Per-tool budgets fix that.
Product
Evals belong in CI, not in a notebook
If an eval only runs when someone remembers to open a notebook, it isn’t a test. We moved ours into the pull request.
Field notes
Field notes: an onboarding agent in a regulated industry
What a logistics team learned shipping an agent that legal had to sign off on line by line.
Engineering
Designing tools agents can actually use
Most agent failures start with a vague tool. Five rules we now apply to every tool schema before it ships.
Company
How we price runs, and why it isn’t per token
Per-token pricing punished the teams doing the right thing. Here is the reasoning behind pricing per run.
Product
What a good trace looks like
A trace should answer “what happened and why” in thirty seconds. The four things we put above the fold.