Work Alongside Scoped Agent Tasks
Continue your own work while Agent Studio runs a separate task. Use distinct working areas and review the agent’s result before accepting it.
Choose tasks with clear ownership, repository boundaries and verification steps. Add parallel work only while you can still understand its dependencies and review the results.
Separate Work That Can Progress Independently
While an agent handles a scoped change, you can work in a separate clone or worktree. A different branch name or terminal alone does not isolate files in a shared checkout.
Example: Plus-One / Plus-N Developer
Suitable tasks might include a documentation correction, a focused test gap or an implementation with a clear interface. Agree on shared dependencies before starting both efforts.
Assign One Reviewable Result
Give the agent an outcome, allowed scope and acceptance criteria. Review its changes for architecture, dependencies and missing checks. Turn remaining issues into concrete tasks.
- Keep your own work in a separate working directory.
- Coordinate changes to shared interfaces.
- Review the result before assigning more dependent work.
Run, Review and Follow Up
Start a task that can be reviewed independently. Continue your own work, then inspect the agent’s result and record your decision. Create follow-up tasks for specific gaps.
Exception example: verdict, run selection and follow-up input on the escalated DEMO-9 task. The capture does not show an active conversation.
- Check changed files, tests and relevant product behavior.
- Include integration conflicts and missing documentation in the review.
- Record why a result was accepted or sent for rework.
Review the Design as Well as the Diff
A successful build answers only part of the review. Inspect component boundaries, duplicated code, ownership and test coverage. Use direct exploration when the product decision is still open; turn settled changes into tasks.
Preserve the Reason for a Change
The task history connects changes to their requests, checks and decisions. A later reviewer can use those records to understand why a change happened and whether the original assumptions still hold.
Activity across projects: a queryable ledger of signals, decisions, and actions with per-project and event-class filters.
Search the task records, logs and repository before asking a model to reconstruct missing history. Save the source links when summarizing earlier work.
Match Parallel Work to Review Capacity
Use this workflow when tasks can be separated and you have time to review each result. Reduce parallelism when work overlaps, requirements are unresolved or unreviewed results accumulate.