This is what the future of coding looks like. Multiple AI models working in the same Codex session, with Jev routing them to tackle different parts of the task. Here, I had it build a full-stack demo app: > Opus 5.5 planned it. > GPT-6 Astra built the backend.... > Kimi K3 built the frontend. > DeepSeek handled testing. > GLM 5.3 Flash wrote the docs. Jev handles the switching. I never touched the model picker. 40% cheaper, 15% faster for the same build.show more

Very interesting usecase. Loved your solution 🙌

how many tokens did opus burn just planning a todo app lol

How do you handle the distributed race conditions even with Jev?

never touching the model picker again sounds great ngl

Looks like I finally found a team that actually listens to my “let's refactor” suggestions. ✨

Matching each task to the model that handles it best sounds like a damn effective approach

Buen ejemplo de enrutar modelos por pedazo de la tarea. En contaduría sirve igual: uno arma la conciliación, otro revisa excepciones. El control sigue en la bitácora y en quien firma el corte.

Splitting one Codex session across planners and builders is the part I want more of. Opus on the plan, Astra on the backend, Kimi on the rest only works if the router actually reads the whole task instead of the first line of the prompt.

That 40% cheaper number is what I'd keep watching over a longer run, because once five models are taking turns you can't assume the cache follows them, and some of those savings can disappear just to keep each model up to speed.

this is exactly what we built it for. one session, the right model on each part, and you never touch the picker. thanks for putting it through a real build

以后写代码像组建临时项目组了娱乐主管

Opus planned it and GPT built it. The builder never had the planner's full context.

this is pretty cool, and it kinda means the overall leaderboard matters less now but what we really needa focus on is which model's best at each piece of the job