20 workers show how reviewing finished agent work opens a reuse loop

Researchers studied 20 workers and more than 10,000 saved task records to see what happens after an agent finishes a job. Most design attention goes to setup and execution. This research focuses on the review stage. It asks how the record of what an agent did can become the raw material for the next job. Post-task records, the structured logs of what an agent did step by step, do real work beyond audit. Workers used them to spot problems, understand what happened, and adapt a completed run into a new one. Reviewing a finished run and building the next one turn out to be the same move. The interface frames them as separate, but the worker treats them as one. The review screen is also where a worker starts over. A worker who can see, annotate, and remix the record of a completed run closes the loop between judging output and starting the next task. Review and starting fresh share the same underlying need: enough structure to see what happened and enough control to act on it.