AI Board workflow
Build mode
Turn several AI models into a coding team: one architect plans and reviews, while worker models implement focused tasks in parallel.
Architect and worker models
In Build mode, the judge model becomes the architect. It breaks your request into concrete tasks, assigns those tasks to worker models, reviews the results, asks for fixes when needed, and writes a final hand-off summary.
Worker models handle focused pieces of the implementation. The architect tracks approvals, fixes, failures, and throughput so the strongest available models are favored over time.
Native Runner V2 kernel
Build execution never runs in the browser. Runner V2 owns durable checkpoints, isolated Git worktrees, native tools, skills, project memory, evidence, provider failover, and recovery. It requires Git and Node.js 24.18.0 or newer and stops before model calls when prerequisites are missing. When the Architect finishes, the run pauses for your final project handoff decision.
When to use it
- Scaffolding a small app or feature from a clear prompt.
- Splitting implementation work across several AI models.
- Letting one model review another model's code before hand-off.
- Running real local checks through a durable native agent kernel.