Customer access becomes more portable.
Discovery, comparison and parts of the purchase can take place outside the provider’s storefront. The provider can gain reach while another system increasingly shapes the options the buyer sees.[3]
ALTERNATE OVERVIEW / STRUCTURAL VIEWSwitch to the partnerships map →
COMMERCE / THE DIVISION OF WORK
Which commerce capabilities do AI platforms still need?
Customer access, purchase execution and control can move separately.
AI partnerships change which company handles each part of a purchase. Some bring customers into AI channels; others bring AI capability into existing businesses.[1]
WAYS THE WORK IS DIVIDEDAI CHANNEL / CAPABILITYTHE CONNECTIONTHE MERCHANT / PROVIDER SYSTEM
ACROSS THESE ARRANGEMENTS
Permission and standards connect the pieces.
Customers authorise purchases; payment providers check credentials; merchants accept orders. A common interface still connects separately operated businesses.[2]
An industry-level synthesis of partnership forms. Bundles show how activities can stay connected; they do not assign identical roles or rights to every case. Select a row to inspect the actual arrangements. Counts overlap across forms and dates.
THE INDUSTRY ARGUMENT
The new boundary is between
forming demand and fulfilling it.
Operations reconnect in different forms.
Some purchases need only a product and an order connection. Others require linked accounts, baskets and fulfilment choices. Companies also use AI providers within their own workflows to improve an existing service.[4]
The connecting role becomes a strategic position.
Platforms, providers and payment systems compete to make their connections useful. Their position depends on participation rules, retained capabilities and the alternatives each side can build.[5]
How comparable is the research across labs?
Records by lab
All held direct records are represented here, but discovery and follow-up work have concentrated on external commerce channels. Google and OpenAI account for 64 of the 81 records.
| Lab coding | Case bundles | Dated records | Documented | Historical | Other stages |
|---|
Each record carries its date and operating stage, including plans, specifications, prototypes and ambiguous rollouts. Lab totals can overlap at case level; the shared multi-lab record is retained separately.
Compare the same questions, not the same number of cases.
Each lab needs a comparable search across customer channels, models inside provider systems, transaction tools, and payment or protocol arrangements. Each relevant configuration then needs the same examination of activity allocation, retained bundles, permissions and change over time. Missing information, an unsearched area and an observed absence must stay distinct.
BACK TO THE PARTNERSHIP LANDSCAPE
See who connects
across these positions.
Return to the partnerships map →Sources and notes
Case evidence. Partnerships Observatory, reconciled Commerce research, cutoff 7 October 2026: 81 dated episode/role records across 58 direct case bundles, including historical, planned and specification material. The displayed record disclosures retain the original source passages, dates and unresolved allocations.
Theory. Carliss Y. Baldwin and Marcus Holgersson, "Ecosystem Evolution: Contracts, Modules, and Standards," forthcoming handbook chapter, held manuscript, pp. 7-11. Compatible modules and design rules can permit replacement. The commerce application remains conditional on working compatibility and usable participation rights.
Third-party analysis. McKinsey, "The agentic commerce opportunity: How AI agents are ushering in a new era for consumers and merchants" (17 October 2025), chapters 1-2, direct and brokered interaction models and interoperability. These distinctions inform the explanation; the Observatory applies them to the arrangements and capabilities in its research.
Theory. Sangeet Paul Choudary, Reshuffle (first edition, July 2025), "Containerization as coordination," pp. 23-24. This informs the distinction between changing a customer interface and coordinating the service behind it. Its application here is the Observatory's interpretation.
Theory. Richard M. Emerson, "Power-Dependence Relations," American Sociological Review 27, no. 1 (1962), pp. 31-41, especially pp. 32-33. Dependence concerns valued outcomes and costly alternatives for the same outcome. The strategic application here is the Observatory's; the analysis examines how alternatives affect negotiating room.