Autopilot / Overview
Overview
Autopilot is the mode where you do the setup and the system does the rest: it writes your activation flows, publishes them, measures them, experiments on them and rewrites them. It stops to ask you exactly once.
What changes compared to building flows yourself
| In the editor | On autopilot | |
|---|---|---|
| Who writes the flow | You do | The system does, from your profile |
| When | When you sit down | Continuously, after setup |
| What the optimizer may do | Patch a flow that exists | Patch, rewrite, and add new flows |
| Where the decision data comes from | Your own traffic | Your traffic, or a pool of similar apps when yours is small |
| What you do | Design, publish, approve each change | Set it up, approve once, read the memo |
The states it moves through
Each app has exactly one autopilot state. Every transition is written to the audit log and appears in the weekly growth memo, so there is never a moment where the system is doing something you cannot name.
App Store listing, SDK, revenue provider, and five questions it has already drafted for you to correct.
Reads your listing and screenshots into an app profile: what the app does, who for, its aha moment, its vocabulary and its visual tokens.
Picks from the activation archetypes, with a reason per flow and a cap on how many. Day one is safe archetypes only.
Generate → apply your theme → render headlessly across devices and languages → repair what broke. Repeats up to a bounded number of rounds.
The only human touch. One screen, one button — and it is only ever reached once.
Publish, activate triggers, set the optimization policy and grant, open the holdout, and roll out on a canary rather than to everyone.
Detect a drop-off, form a hypothesis, validate the patch against your constraints, run the experiment, promote what wins, write what it learned to the pattern library.
When a flow runs out of winning patches, or a new gap appears in the profile, it authors a fresh candidate and returns to AUTHORING.
The approval gate is passed once
This surprises people, so it is worth stating plainly: after you approve the first set, later authoring rounds do not come back for approval. That is what “autopilot” was decided to mean. What protects you afterwards is not a second gate but the machinery around it — headless validation before publish, canary rollout, guardrails that roll back on their own, the locks you set at the gate, and the kill switch.
One exception is structural rather than procedural: when autopilot rewrites a flow, the new version does not replace the old one. It enters as an experiment arm against the flow it wants to supersede, and only takes over if it wins.
What it will not do
- Touch your own screens.Autopilot changes Uplift flows and nothing else. Your app's UI, purchase code, sign-in and permission prompts are yours.
- Write your paywall on day one. The paywall and the trial-lifecycle flow wait for data about your app. What it writes explains why.
- Claim a result it cannot measure. Every number carries an evidence level; unreadable is reported as unreadable, never as zero. See Evidence & claims.
- Send messages to your users. No push, no email. It renders flows inside your app.
Next
Setup is the only part that needs you. If you would rather see the machinery first, start with Safety & guardrails.