NickFORGE

Thanks for calling that out — fair question. The 85–92% number is from internal time-tracked tests comparing:

Before FORGE (baseline):

Spec/plan creation: 3–6 hours

Context switching (browser → IDE → terminal): ~1 hour

Manual copy/paste of generated code: 30–60 minutes

Manual setup + environment config: 30–90 minutes

After FORGE (CloudShell + agents):

Plan generation: 10–30 seconds

No local setup

Agents edit files directly in the repo

Terminal + UI in one interface

So the measurement isn’t “marketing hype” — it’s based on reducing a 6–12 hour workflow to 30–60 minutes on real feature builds.

Happy to share the breakdown or the feature sets we used if that helps.

Thanks - glad it’s useful. The split screen came from our own frustration switching between a plan and the agent actually editing files. If anything feels off or confusing as you try it, I’m around and happy to adjust things.

We’ve been experimenting with a similar idea but in a browser-native environment — running real containers + a WebSocket terminal + multi-agent workflows. GPT-5.1 (Codex Max especially) seems to handle multi-step refactors a lot more cleanly, and chaining it through CLI agents has been surprisingly reliable.

Curious if anyone else is trying agent orchestration beyond the editor itself?