Ship an agent-built dictation app through cross-model review and voice tests
At a glance
A solo Mac app builder used separate building, review and QA agents, with his own voice test as the final release gate. This workflow is listed under coding agents, code review, release testing.
- Agents / tools
- Claude, GPT-5.5, Codex, Git worktrees, Clabrate
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Prepare the app repository and access to the coding models. Put concurrent feature work in separate Git worktrees and tell each agent that other agents are working too. 2) Specify the change and architectural requirements in writing, then assign implementation to a builder agent. 3) Have GPT-5.5 and Claude review the same diff adversarially and run a separate QA gate; inspect the diffs yourself. 4) Build the app and test the actual dictation behavior with your own voice. Count the change as finished only after that live test, then ship the tested build.
How it went
The author reported shipping Clabrate v0.6.0 after 31 active days across seven weeks, writing none of the code himself. He reported only a handful of downloads and no sales.
Worth knowing
Agents initially reverted each other's changes in a shared checkout. His $13,500 figure was estimated API list-price usage; he paid through subscriptions, not that amount in API charges.
