STEP 01
WHAT WE TAKE IN
Describe the workflow
State the job in plain language. The risk threshold. The handoff. Autopilot turns the brief into a structured spec.
→AI LABS · AUTOPILOT · DESCRIBE IT. WATCH IT RUN.
Describe it. Watch it run.
Workflows that build themselves — and wait for the approval you set. Describe the job. Inspect the plan. Set the gates. The workflow runs on your approval, with the record attached.
DESCRIBE
Plain language in
State the job, the risk, the handoff — once.
APPROVE
Plan before run
Gates, reviewers, evidence — set before launch.
RUN
With proof attached
Same path. Same record. Every time.
WORKFLOW READOUT
The plan, before it runs.
Signal in, review at the gate, approval out. Every node is named. Every gate is owned. Nothing runs until the plan is signed.
HOW IT WORKS
Three steps. No copied playbook.
Describe the workflow. Approve the plan. Watch it run.
STEP 02
WHAT WE PUT UP FOR REVIEW
Approve the plan
The plan shows the steps, the reviewers, the escalations, and the record it will leave. You approve before anything runs.
→STEP 03
WHAT WE EXECUTE
Watch it run
The approved workflow runs on the same gates, the same ownership, the same proof — every time it is used.
WHAT COMES OUT
What your team leaves with.
Every run leaves something the team can act on — and something the next launch has to clear.
01
Workflow specs
The structured brief — task, risk posture, handoffs, evidence rules — written down before anything runs.
↳ ARTIFACT
02
Approval gates
Named reviewers. Named thresholds. Named escalations. The plan everyone signs before launch.
↳ ARTIFACT
03
Run logs
Every node, every retry, every guardrail trip. The workflow's own readout, written as it runs.
↳ ARTIFACT
04
Signed run records
The approved plan, the inputs, the outputs, and the seal — ready when someone asks who approved what.
↳ ARTIFACT
WHERE IT FITS
In the loop, this is where you test.
Autopilot is how the run gets described, planned, and put up for review. Test the run. Review the hard cases. Recruit the right specialist. Remember the misses. Approve what's right.
01
Test
● YOU ARE HERE
02
Review
03
Recruit
04
Remember
05
Approve
RELATED MODULES
Next to this in the Model OS.
TRAINING
The training run, written down.
Every run a record. Every record a fact.
See the page →RL ENVIRONMENTS
Environments that hold still.
Deterministic worlds so the score means the same thing twice.
See the page →SYNTHETIC POPULATIONS
The cases you don't have, yet.
Populations built to find the failure before the customer does.
See the page →AUTOPILOT
Describe it. Watch it run.
Bring the workflow your team still rebuilds by hand. We'll turn it into a plan the team signs — and a run that leaves the record.