How we keep every run under two seconds

A workflow that takes a minute to run feels broken, even when it is correct. So we gave every run a budget: two seconds from trigger to finished action.

Do the slow work once

Most of a run’s time used to go into looking things up: who owns a region, which pipeline a deal belongs to. We now cache that context per workspace and refresh it in the background.

Fail fast, then retry

When a tool is slow to answer, the run pauses and retries on its own instead of blocking everything behind it. You see the delay in run history, and nothing is lost.

What it means for you

Today the average run finishes in under a second, and 99% finish inside the two-second budget.