Jethro shares today's build-in-public update: testing a "gauntlet loop" approach for AI-assisted game development after hearing about similar workflows being used to create Call of Duty-style projects. He started the project in Claude Code, hit token limits, and then moved work over to Grok and Codex subagents to keep the build moving.
The big takeaway is that AI coding is powerful, but it still is not truly hands-off. Even with a PRD and a hoped-for one-shot workflow, the process required repeated check-ins, course correction, and supervision. The game is now running in the simulator, with the next step being a real-device test on Jethro's phone.
0:00 — Starting a gauntlet-loop app build
0:17 — Running into Claude token limits
0:31 — Switching to Grok and Codex subagents
0:53 — Why one-shot AI coding still needs supervision
1:28 — Keeping the AI responsible for the changes
1:58 — Simulator success and next device test
Project: AI-assisted iPhone and iPad game build
Tools mentioned: Claude Code, Grok, Codex, subagents
Key lesson: A detailed PRD helps, but AI coding still needs active oversight
Next step: Install and test the game on a real phone