Type: url
Original: https://developers.cyberagent.co.jp/blog/archives/62110/
Status: generated
Summary
A two-engineer team at CyberAgent describes an AI-driven development setup in which AI leads requirements definition, technical design, task splitting, implementation and review, while the humans only make decisions and confirm output. The repository holds 37 Skills and 24 SubAgents, with orchestrator Skills launching specialized SubAgents per phase, and review gates that auto-approve a pull request once both a technical and a specification review score 80 or above. The author argues the gains come from supervising several parallel tasks rather than from any single task finishing faster, and names undocumented tacit knowledge and the human limit on reviewing AI output volume as the remaining obstacles.