Bring the work.
Keep the decision.
Two ways to put ML.ai Code to work: build a change, or plan it before touching the code.
“Make checkout
safe to submit.”
From the request
to a reviewed change.
Let lighter model routes handle reading and planning. Reserve stronger reasoning for the proposed change and its checks.
async function handlePay() {- await charge(); if (pending.current) return; pending.current = true; try { await charge({ idempotencyKey }); } finally { pending.current = false; }}Understand
Inspect the repository and establish the work.
Prepare
Use the appropriate agent and model for the step.
Review
See the proposed change before allowing a write.
See the path.
Before changing the code.
Architect identifies the relevant files, trade-offs and sequence of work without changing the code. Plan can turn that reasoning into a file for your review.
A clear plan is a useful outcome on its own.
A safer payment flow
PayButton.tsx → charges.ts
Server-side idempotency
Double-submit regression
The editor is just
one place to start.
ML.ai Code brings ML.ai into the editor. Explore the orchestration layer when the question moves from one task to a workload.
Build and Planning are the two workflows on ML.ai’s use-case page. The checkout scenario above is a visual example, not a customer result.
Bring your next task.
Keep the final say.
Start in the editor, or explore the ML.ai orchestration behind it.