HN-FQ-001 · Living question · cutoff 2026-11-02

If AI agents do the work, who owns the customer?

By 2028-12-31, when an AI agent materially shapes a consumer purchase, will the merchant, the agent platform, or a shared protocol layer usually control the continuing customer relationship?

OpenNo overall numerical probabilityInitial entry only

Provisional answer

The relationship is most likely to split by surface. Merchants retain the transaction, fulfillment, and often the account when the agent hands the buyer back to a branded storefront. Agent platforms are gaining discovery, comparison, and conversational intent. Protocol and payment layers can make the handoff auditable without deciding who owns the relationship. The first useful forecast is therefore about control points and evidence, not a winner-takes-all probability.

No overall probability is issued. This is a qualitative research inquiry with coexisting scenarios for comparison; no single scored winner is issued or settled. The current evidence is mostly platform and infrastructure commitments rather than a held-out adoption dataset.

Competing futures

Different paths, different evidence.

These scenarios are non-exhaustive; probabilities are intentionally omitted.

MERCHANT_FRONT_DOORNo probability issued

Merchant keeps the front door

Agents become a discovery and comparison layer, but shoppers usually finish inside merchant-controlled accounts, storefronts, loyalty systems, and service channels.

What separates it: Branded checkout, account linking, loyalty, post-purchase service, and merchant-owned permission records remain the dominant path after discovery.

PLATFORM_GATEKEEPERNo probability issued

Agent platform becomes the gatekeeper

A small number of consumer agent surfaces own the repeated conversation, ranking, memory, and re-engagement while merchants compete to be selected inside that interface.

What separates it: Consumers return to the agent for repeat shopping, merchant exposure depends on platform ranking or feed access, and post-purchase interaction stays inside the agent surface.

PROTOCOL_SHAREDNo probability issued

Control is shared by protocol

Open commerce and identity protocols let merchants, agents, payment providers, and consumers retain distinct rights, with receipts and consent records making the relationship portable.

What separates it: A buyer can move between agents without losing merchant identity, permissions, receipts, or service history, and merchants can switch agent channels without rebuilding the transaction layer.

What is tracked

Watch the handoff.

  • Whether agent shopping flows return buyers to a merchant account or keep service inside the agent interface.
  • Whether merchant protocols expose customer identity, loyalty, consent, and post-purchase state as portable records.
  • Whether merchants can measure repeat relationships across agent referrals rather than only receiving a completed order.
  • Whether consumers can override, switch, or revoke an agent without losing access to a merchant relationship.

What is not tracked

Keep the boundary.

  • A claim that any company legally owns a person or their data based on marketing language.
  • Private platform usage numbers or vendor badges that cannot be independently checked.
  • A general forecast of AI capability, retail sales, or the future of work.
  • A reliability claim about any agent, payment network, or merchant integration.

Leading indicators

What would move the read.

  • Merchant integrations that expose account linking, loyalty, fulfillment, and service state inside agent experiences.
  • A repeat-purchase metric that attributes the continuing relationship, not only the first referral or checkout.
  • Agent protocols that publish consent, identity, receipt, and revocation semantics across more than one platform.
  • Consumer research or regulator guidance that tests control, switching, and post-purchase recourse in agent-mediated buying.

Failure signals

What could make this wrong.

  • A broad independent measurement shows that agent-mediated purchases mostly end in merchant-owned accounts and service channels through 2028.
  • Consumers consistently refuse agent-held post-purchase interaction even while using agents for product discovery.
  • Merchant integrations remain too fragmented for any agent surface to retain repeat intent or service context.
  • A portable protocol becomes the default record for identity, consent, receipts, and switching across major agent and merchant systems.

Next review conditions

Return with a reason.

  • 2027-03-31: review whether major agent shopping launches publish account, loyalty, service, and attribution behavior rather than only catalog or checkout behavior.
  • 2027-09-30: review at least one independent measurement of repeat purchase, customer-service handoff, or switching in an agent-mediated flow.
  • 2028-06-30: update the scenario differentiation table if a portable identity, consent, or receipt protocol has measurable cross-platform use.
  • 2028-12-31: resolve against the documented sample and publish the evidence boundary, including missing data.

Record history

  • 2026-11-02 · initialInitial assessment created from six public primary sources checked on 2026-11-02. The evidence supports a split-control hypothesis, not a winner, and no historical updates are implied.Sources: SRC-001, SRC-002, SRC-003, SRC-004, SRC-005, SRC-006

Evidence ledger

Primary sources, dated checks.

Source freshness describes this review, not the truth of a claim. Company pages show stated designs or deployments; they do not establish adoption on their own.

  1. Google: Under the Hood — Universal Commerce Protocolofficial protocol · published 2026-01-11 · checked 2026-11-02 · currentProtocol design and Google’s stated merchant-of-record, business-logic, and consent-proof model.
  2. OpenAI: Powering Product Discovery in ChatGPTofficial product · published 2026-03-24 · checked 2026-11-02 · currentProduct evidence that discovery is moving into ChatGPT while merchants retain options for their own checkout and branded experiences.
  3. OpenAI Developers: Agentic Commerce Protocolofficial protocol · published undated page · checked 2026-11-02 · currentProtocol documentation map; not the discovery-scoped connective-layer wording.
  4. OpenAI Developers: Agentic Commerce key conceptsofficial protocol · published undated page · checked 2026-11-02 · currentTechnical separation between the ChatGPT-rendered checkout experience and merchant-controlled checkout state, tax, risk, payment, and order acceptance.
  5. Visa: Agentic commerceofficial company update · published undated page · checked 2026-11-02 · currentMerchant-side framing of agent discovery, merchant assistants, customer relationship control, and agent-friendly checkout.
  6. Visa: First live pilot for trusted AI agent paymentsofficial company update · published 2026-10-20 · checked 2026-11-02 · currentDeployment evidence for approved B2B agent payments with buyer controls and supplier oversight; it is not evidence of broad consumer adoption.