Definition
Autonomous commerce describes a spectrum of decision-making authority in a transaction, running from assisted through delegated to fully autonomous, where an AI completes a purchase with no per-transaction human review. Assisted means an AI recommends and a human clicks buy; delegated means an AI acts within pre-set limits. In general use the term points at the far end of that scale — systems that decide and transact with minimal human involvement — but it is only workable as a scale, because each level demands different governance. It is a framing of how much authority is handed to the system, independent of whether the purchase itself involves any clicks at all.
Key points
- It is a spectrum, not a binary: assisted, delegated, and fully autonomous are three distinct points along it.
- Each level demands different governance: spend limits, category restrictions, and return-handling authority all scale with the level of delegation.
- It is measured by decision authority, not by interface friction — that is what separates it from zero-click commerce.
- A transaction can be zero-click but still assisted, or fully autonomous but still involve a click to arm the policy.
What does each level of autonomy require in practice?
Assisted systems need nothing beyond a recommendation surface — the human makes the final call. Delegated systems need a spend cap per transaction or per period, a category allowlist limiting what the agent can buy, and an escalation threshold that routes unusual purchases back to a human. Fully autonomous systems need all of that plus a defined return and dispute path, since no human reviewed the original choice and the buyer needs recourse if it was wrong. An audit log of what the agent decided and why is the artifact that makes any of these levels accountable after the fact.
Common pitfalls
- Granting full autonomy with no spend cap or category restriction, so a single bad inference has no ceiling on its cost.
- No defined return or dispute path for autonomous purchases, leaving the buyer without recourse when the agent chose wrong.
- Treating "delegated" and "fully autonomous" as the same thing, when the governance each requires is materially different.
FAQ
How is this different from zero-click commerce?
Zero-click commerce measures interface friction — whether a storefront visit or click-through happened. Autonomous commerce measures decision authority — how much of the choice a human reviewed. The two overlap often but are not the same axis: a purchase can be zero-click and still human-approved, or fully autonomous and still require a click to arm the policy.
What decides which level a given purchase should sit at?
The cost of being wrong, which is a function of unit value and reversibility. Low-value, freely returnable, repeatedly bought consumables tolerate delegation cheaply. Configured, regulated, perishable, or non-returnable items keep their decision assisted, because there is no affordable way to undo a bad autonomous choice on them.
Source
No published standard defines levels of purchasing autonomy, so the nearest documented material describes the surfaces where delegation currently stops rather than the spectrum itself. Google's documentation on AI features in Search says that AI Overviews and AI Mode "surface relevant links to help people find the information they're looking for", that AI Mode is most useful for queries needing further exploration, reasoning, or complex comparisons, and that appearing in either requires no new machine-readable files and no special schema.org structured data. It describes an assistive surface that hands a person links to act on; it never mentions commerce, agents, or purchasing at all. That makes it a useful marker for the assisted end of this spectrum and no guide whatever to the delegated and fully autonomous ends. This page publishes its own definition as a schema.org DefinedTerm, the vocabulary used in its JSON-LD, so the definition can be extracted without the surrounding prose.