Run a team of 28 coding agents to rebuild an app designer
GoodBarber's CTO ran 28 coding agents, up to 20 at once, to restructure an app designer's code; he reports 569 pieces handled in six days, with 0 gaps across 12,074 test cases.
Done with OtherClaude Opus 5.5
What they did
GoodBarber's CTO spent two weeks planning the work with Claude Opus 5.5 and about 200 helper agents. Then he let 28 worker agents run on their own, up to 20 at once. One coordinator agent alone could combine and publish changes. A shared rules file, a task queue and a progress screen kept them in step. Checks compared the app before and after each change, including undo and redo.
How it went
He reports 569 pieces finished in six days: 445 done and 124 dropped or merged. A final check of 12,074 test cases found 0 gaps. He followed the agents' advice on 47 of 59 questions, and accepted 166 of 185 earlier decisions while rejecting 19. One overnight run finished 173 pieces in 11 hours.
Worth knowing
The human cost was decision fatigue, not hours. Questions arrived all day and piled up, once 63 in an hour. Ten of the 19 rejected decisions involved an agent leaving out something that was hard to handle.
Try it yourself
Help me plan a parallel agent refactor of [codebase/module]. First, propose how to split the work into independent units and define 'done' as automated equivalence checks: before/after state snapshots, an exact undo, and parity between the API and the UI. Design a queue in which workers each use their own branch and worktree, plus one integrator that batches merges to [dev environment]. When a worker hits a decision it cannot make alone, it should post a question with 2-3 options, their costs and a recommendation, then move on to the next unit. Show me the plan and the protocol file, and wait for my approval before running any workers or deploying.
Swap the highlighted [parts] for your own details.
Source: DEV Community · Dominique Siacci · Oct 15, 2026
Written with AI assistance from the original post and checked against our editorial guidelines. Results are as reported by the source, not independently verified.
