Autonomy is the single setting that decides whether BoostEcom is a copilot or an operator. It is set per store, and a new store starts at level 1.

| Level | What the AI does | |---|---| | 1 | Proposes actions. You approve every one. | | 2 | Executes, but each execution requires validation. | | 3 | Executes autonomously within its remit. | | 4 | Runs the business end to end. |

What raising the level actually changes

It changes who confirms, and nothing else. In particular it does not change:

The Shopify scopes. Those are approved once, on Shopify's consent screen or in your Custom App. Level 4 with read_products only still cannot write a product.
The credit guardrails. A level 4 store hits the same 402 refusal and the same mid-stream abort as a level 1 store.
The audit trail. Every action an agent takes is written to AuditLog at every level. Autonomy removes a confirmation step, never a record.

Choosing a level

The honest guidance, rather than a push up the ladder:

Level 1 while you are still learning what the agents propose. The proposals themselves are the value at this stage: you are calibrating trust.
Level 2 once the proposals are consistently right and the bottleneck is you retyping the same approval.
Level 3 per remit, not globally. Letting @Maya reword meta descriptions autonomously is a different risk from letting @Marco reprice a catalogue.
Level 4 for stores where the downside of a wrong action is bounded and you have a rollback path.

Where the platform is

Autonomy levels describe the product roadmap as much as a setting:

| Phase | State | |---|---| | Phase 1 — AI Copilot | Current. | | Phase 2 — AI Operator | Executes operations. | | Phase 3 — Autonomous businesses | Target. |

Levels 3 and 4 are reachable today for the surfaces the platform already writes to. They are not a blanket switch that makes every agent autonomous over every surface, and any page that told you otherwise was overselling.