Agentic commerce has moved from concept to infrastructure and early production. AI interfaces can already discover products, create carts and execute authenticated payments; open standards now cover commerce, payments and agent identity. The market is still pre-scale because trust, consent, liability, fraud, interoperability and merchant integration remain unresolved. The key transition is from “can an agent pay?” to “can every participant prove who the agent is, whose intent it carries, what it was authorised to do, and who is accountable when something goes wrong?”
Research question and scope
Core question: How mature is agentic commerce in 2026, what technical/economic stack is emerging, and what must merchants, platforms and payment providers do before delegated buying scales?
Scope: Consumer and B2B agentic commerce, product discovery/checkout protocols, payment protocols, agent identity, live transactions, machine-to-machine payments and consumer trust. It does not assume fully autonomous purchasing is already mainstream.
Method: Uses live platform launches, developer specifications, payment-network protocols, production transaction evidence, IMF analysis and consumer trust research. Announced frameworks are distinguished from deployed standards.
1. Agentic commerce is now a real stack, not a single product
By 2026, multiple layers have emerged: OpenAI and Stripe’s Agentic Commerce Protocol (ACP) for shopping/checkout; Google’s Universal Commerce Protocol (UCP) for interoperable commerce journeys; Google’s Agent Payments Protocol (AP2), now donated to the FIDO Alliance, for secure agent-initiated payments; Visa’s Trusted Agent Protocol for merchant recognition; Mastercard’s Agent Pay and Verifiable Intent; and newer machine-payment protocols for agent-to-service transactions. S1S2S3S4S5S6
| Layer | Representative standard/service | Problem solved |
|---|---|---|
| Discovery/catalogue | ACP product discovery; UCP | Expose structured products, availability, terms and checkout capabilities |
| Commerce workflow | ACP; UCP | Create cart/order and exchange required commerce state |
| Payment authorisation | AP2; Mastercard Verifiable Intent | Prove user-authorised intent and payment scope |
| Agent identity | Visa Trusted Agent Protocol; Mastercard credentials; KYA initiative | Distinguish legitimate agents from bots and link activity to accountable parties |
| Payment execution | Agent Pay, card/account rails, shared tokens | Execute authenticated transaction without exposing raw credentials |
| Machine economy | Stripe/Tempo MPP; Mastercard Agent Pay for Machines | High-frequency, programmatic agent-to-service payments |
2. The shopping interface has already moved inside AI
OpenAI introduced Instant Checkout in ChatGPT in September 2025 using ACP, initially enabling direct purchase from US Etsy sellers and announcing Shopify merchant support. In March 2026, it expanded ACP into richer product discovery and comparison. S1S7 Stripe’s March 2026 field guide reports live work with merchants including Etsy and URBN and describes agentic checkout as a production integration problem rather than a demonstration. S8
3. Live payments establish technical feasibility
Mastercard, Worldline and ING completed a live European end-to-end agentic payment in production in June 2026, following other authenticated agentic payment demonstrations. S9 Visa and Artemis’ July 2026 analysis uses live onchain data to examine agents paying for compute, data and other services. S10 These examples establish that payment execution is technically feasible; they do not establish consumer-scale adoption.
4. Trust, identity and intent are the control plane
Visa’s Trusted Agent Protocol uses cryptographic signatures so merchants can verify that an incoming automated request comes from a recognised agent and carries commerce intent. S4 Mastercard’s Verifiable Intent framework records consumer identity, original instructions and transaction outcome in a tamper-resistant chain, with selective disclosure. S11
The latest market direction is toward interoperability. On 10 September 2026, Ant International, Mastercard and Visa announced collaboration on a Know-Your-Agent framework intended to let networks, wallets, agent platforms and marketplaces recognise trusted agents across ecosystems while retaining their own risk decisioning. S12 This is an initiative under development rather than a finished universal standard, but it is a strong signal that agent identity is becoming an industry-level requirement.
- Who is the agent and who operates it?
- Which human or business authorised the agent?
- What exact intent, limit, merchant/category and time window were authorised?
- Can the merchant prove that the request was not modified?
- Which credential may be used and only for what transaction?
- What happens if the agent chooses the wrong product or exceeds intent?
- Who handles refunds, disputes, returns, fraud and customer support?
Sources and notes
- OpenAI, Instant Checkout and Agentic Commerce Protocol (29 Sep 2025). [Source](https://openai.com/index/buy-it-in-chatgpt/)) — Live in-context commerce launch and ACP architecture.
- Google, Universal Commerce Protocol (11 Jan 2026). [Source](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/)) — Open-source commerce standard with retailer/payment partners.
- Google, AP2 donation to FIDO Alliance / v0.2 (28 Apr 2026). [Source](https://blog.google/products-and-platforms/platforms/google-pay/agent-payments-protocol-fido-alliance/)) — Human-not-present payments and Verifiable Intent interoperability.
- Visa, Trusted Agent Protocol — developer specification. [Source](https://developer.visa.com/capabilities/trusted-agent-protocol/docs)) — Cryptographic agent recognition and merchant implementation.
- Mastercard, Agent Pay. [Source](https://www.mastercard.com/us/en/business/artificial-intelligence/mastercard-agent-pay.html)) — Network-level agent credentials, intent and payment infrastructure.
- Stripe, Machine Payments Protocol (18 Mar 2026). [Source](https://stripe.com/blog/machine-payments-protocol)) — Open standard for agent/machine payments.
- OpenAI, Powering Product Discovery in ChatGPT (24 Mar 2026). [Source](https://openai.com/index/powering-product-discovery-in-chatgpt/)) — ACP expansion from checkout into product discovery.
- Stripe, How to prepare for agentic commerce: technical field guide (10 Mar 2026). [Source](https://stripe.com/guides/how-to-prepare-for-agentic-commerce-technical-field-guide)) — Production integration lessons across merchants.
- Mastercard + Worldline + ING, live European agentic payment (2 Jun 2026). [Source](https://www.mastercard.com/news/europe/en/newsroom/press-releases/en/2026/worldline-ing-and-mastercard-complete-a-live-end-to-end-european-agentic-payment-in-production/)) — Live production payment evidence.
- Visa + Artemis, Agentic Payments: What Onchain Data Reveals (14 Jul 2026). [Source](https://www.visa.com/en-us/thought-leadership/innovation/agentic-payments-from-the-ground-up)) — Live onchain agent-payment evidence.
- Mastercard, Verifiable Intent framework explainer (2026). [Source](https://www.mastercard.com/content/dam/mccom/eu/news-and-trends/business-thought-leadership/digital-payments/pdfs/Get_ready_for_AI-powered_shopping_experiences.pdf)) — Cryptographic intent, identity and privacy model.
- Reuters, Visa/Mastercard/Ant International Know-Your-Agent initiative (10 Sep 2026). [Source](https://www.reuters.com/technology/payment-firms-visa-mastercard-ant-international-team-up-ai-agent-trust-framework-2026-09-10/)) — Latest interoperability initiative; framework under development.
8i TECHNOLOGIES
Agentic Commerce
From AI-assisted shopping to delegated transactions: protocols, payments, trust, merchant readiness and the emerging machine economy
Reference date: 15 September 2026
See what matters next.
| Executive answer Agentic commerce has moved from concept to infrastructure and early production. AI interfaces can already discover products, create carts and execute authenticated payments; open standards now cover commerce, payments and agent identity. The market is still pre-scale because trust, consent, liability, fraud, interoperability and merchant integration remain unresolved. The key transition is from “can an agent pay?” to “can every participant prove who the agent is, whose intent it carries, what it was authorised to do, and who is accountable when something goes wrong?” |
|---|
Research question and scope
Core question: How mature is agentic commerce in 2026, what technical/economic stack is emerging, and what must merchants, platforms and payment providers do before delegated buying scales?
Scope: Consumer and B2B agentic commerce, product discovery/checkout protocols, payment protocols, agent identity, live transactions, machine-to-machine payments and consumer trust. It does not assume fully autonomous purchasing is already mainstream.
Method: Uses live platform launches, developer specifications, payment-network protocols, production transaction evidence, IMF analysis and consumer trust research. Announced frameworks are distinguished from deployed standards.
1. Agentic commerce is now a real stack, not a single product
By 2026, multiple layers have emerged: OpenAI and Stripe’s Agentic Commerce Protocol (ACP) for shopping/checkout; Google’s Universal Commerce Protocol (UCP) for interoperable commerce journeys; Google’s Agent Payments Protocol (AP2), now donated to the FIDO Alliance, for secure agent-initiated payments; Visa’s Trusted Agent Protocol for merchant recognition; Mastercard’s Agent Pay and Verifiable Intent; and newer machine-payment protocols for agent-to-service transactions. [S1][S2][S3][S4][S5][S6]
| Layer | Representative standard/service | Problem solved |
|---|---|---|
| Discovery/catalogue | ACP product discovery; UCP | Expose structured products, availability, terms and checkout capabilities |
| Commerce workflow | ACP; UCP | Create cart/order and exchange required commerce state |
| Payment authorisation | AP2; Mastercard Verifiable Intent | Prove user-authorised intent and payment scope |
| Agent identity | Visa Trusted Agent Protocol; Mastercard credentials; KYA initiative | Distinguish legitimate agents from bots and link activity to accountable parties |
| Payment execution | Agent Pay, card/account rails, shared tokens | Execute authenticated transaction without exposing raw credentials |
| Machine economy | Stripe/Tempo MPP; Mastercard Agent Pay for Machines | High-frequency, programmatic agent-to-service payments |
2. The shopping interface has already moved inside AI
OpenAI introduced Instant Checkout in ChatGPT in September 2025 using ACP, initially enabling direct purchase from US Etsy sellers and announcing Shopify merchant support. In March 2026, it expanded ACP into richer product discovery and comparison. [S1][S7] Stripe’s March 2026 field guide reports live work with merchants including Etsy and URBN and describes agentic checkout as a production integration problem rather than a demonstration. [S8]
3. Live payments establish technical feasibility
Mastercard, Worldline and ING completed a live European end-to-end agentic payment in production in June 2026, following other authenticated agentic payment demonstrations. [S9] Visa and Artemis’ July 2026 analysis uses live onchain data to examine agents paying for compute, data and other services. [S10] These examples establish that payment execution is technically feasible; they do not establish consumer-scale adoption.
4. Trust, identity and intent are the control plane
Visa’s Trusted Agent Protocol uses cryptographic signatures so merchants can verify that an incoming automated request comes from a recognised agent and carries commerce intent. [S4] Mastercard’s Verifiable Intent framework records consumer identity, original instructions and transaction outcome in a tamper-resistant chain, with selective disclosure. [S11]
The latest market direction is toward interoperability. On 10 September 2026, Ant International, Mastercard and Visa announced collaboration on a Know-Your-Agent framework intended to let networks, wallets, agent platforms and marketplaces recognise trusted agents across ecosystems while retaining their own risk decisioning. [S12] This is an initiative under development rather than a finished universal standard, but it is a strong signal that agent identity is becoming an industry-level requirement.
- Who is the agent and who operates it?
- Which human or business authorised the agent?
- What exact intent, limit, merchant/category and time window were authorised?
- Can the merchant prove that the request was not modified?
- Which credential may be used and only for what transaction?
- What happens if the agent chooses the wrong product or exceeds intent?
- Who handles refunds, disputes, returns, fraud and customer support?
5. Consumer willingness is behind technical capability
Visa’s September 2026 Trust Index, conducted by Harris Poll among 2,065 US consumers, found 72% had used an AI assistant but only 23% trusted generative AI to handle payment transactions on their behalf. Trust rose to 61% when Visa was the payment brand securing the transaction. [S13] The finding is US-only and brand-specific, but it makes the adoption bottleneck clear: autonomous transaction capability requires inherited trust from payment, identity and merchant systems.
6. Human-present and human-not-present are different markets
| Mode | Example | Risk profile | Likely adoption |
|---|---|---|---|
| AI-assisted / human confirms | Agent recommends; user approves checkout | Comparable to advanced digital checkout | Scaling now |
| Delegated within limits | Agent may purchase below price/category threshold | Intent, budget and item correctness | Next phase |
| Human-not-present | Agent acts when condition occurs | Timing, fraud, changed context, revocation | Early; AP2 v0.2 explicitly targets this |
| Machine-to-machine | Agent purchases compute/data/services | Identity, micro-payment economics, automated disputes | Emerging B2B/infrastructure market |
Google’s April 2026 AP2 v0.2 explicitly added support for “Human Not Present” payments based on pre-authorised instructions, illustrating the shift from confirmation-centric checkout toward bounded delegation. [S3]
7. Merchant readiness is more than payments
Stripe’s production lessons emphasise that real commerce involves catalogues, inventory, taxes, fulfilment, returns, fraud systems and customer support—not simply a payment API. [S14] Agentic commerce therefore requires merchants to expose reliable machine-readable product and policy state while keeping their existing merchant-of-record responsibilities.
| Merchant capability | Why agents need it |
|---|---|
| Structured/current catalogue | Compare products accurately and know what can be bought now |
| Machine-readable policy | Evaluate shipping, return, eligibility and warranty constraints |
| Inventory/price freshness | Avoid recommending or ordering stale offers |
| Agent recognition | Separate valuable agent traffic from scraping/fraud bots |
| Scoped payment credentials | Pay without exposing reusable card/account data |
| Order/fulfilment APIs | Track, amend, cancel and support the transaction |
| Evidence/claims | Allow agents to justify recommendations and meet regulated requirements |
8. Protocol proliferation is a feature of an early market—and a risk
ACP, UCP, AP2, Visa TAP, Mastercard frameworks and machine-payment protocols address overlapping but different layers. The risk for merchants is one-off integration to every agent and network. Google’s UCP is explicitly designed to interoperate with AP2, Stripe positions its Agentic Commerce Suite as a compatibility layer across protocols, and the September 2026 Know-Your-Agent initiative aims at cross-network identity recognition. [S2][S8][S12] The direction is toward composable standards, but convergence is not complete.
9. B2B and machine commerce may scale differently from consumer shopping
The consumer story receives attention, but machine-to-machine commerce could be structurally important because software agents already consume metered APIs, compute and data. Stripe’s Machine Payments Protocol and Mastercard Agent Pay for Machines both target continuous, programmatic transactions that can be small, frequent and non-human-initiated. [S6][S15] This creates new pricing models for data, software capabilities and machine services.
10. 8i maturity model
| Stage | Customer/agent role | Infrastructure status | 8i assessment |
|---|---|---|---|
| 1 — Recommend | Agent discovers/compares; human buys | Mature AI discovery | USE |
| 2 — In-context checkout | Agent carries cart/checkout; human confirms | Live on major platforms | TEST / SCALE |
| 3 — Delegated purchase | Agent transacts within explicit limits | Standards and pilots emerging | TEST |
| 4 — Human-not-present | Agent executes when condition met | Early protocol support | WATCH |
| 5 — Machine economy | Agents buy from agents/services continuously | Early infrastructure | WATCH |
11. Strategic implications
- Merchants should become agent-readable before they attempt autonomous checkout.
- Payment providers and identity networks can gain strategic importance as trust infrastructure.
- Platforms that own customer intent may gain distribution power even if merchants remain merchant of record.
- Agent identity/KYA will become a new fraud and access-control layer alongside KYC and merchant verification.
- Product/price/service information becomes an API asset, not only web content.
- Liability and dispute models need to distinguish model error, agent-policy failure, payment fraud and merchant fulfilment failure.
- Machine-to-machine microtransactions could create markets for data and services that are uneconomic under human checkout flows.
Source register
Sources are prioritised toward primary institutions, transparent methodology, observable behavioural data, large-sample surveys and current production evidence. Vendor evidence is identified as such and used cautiously where it provides proprietary usage data or live implementation detail.
[S1] OpenAI, Instant Checkout and Agentic Commerce Protocol (29 Sep 2025). Source — Live in-context commerce launch and ACP architecture.
[S2] Google, Universal Commerce Protocol (11 Jan 2026). Source — Open-source commerce standard with retailer/payment partners.
[S3] Google, AP2 donation to FIDO Alliance / v0.2 (28 Apr 2026). Source — Human-not-present payments and Verifiable Intent interoperability.
[S4] Visa, Trusted Agent Protocol — developer specification. Source — Cryptographic agent recognition and merchant implementation.
[S5] Mastercard, Agent Pay. Source — Network-level agent credentials, intent and payment infrastructure.
[S6] Stripe, Machine Payments Protocol (18 Mar 2026). Source — Open standard for agent/machine payments.
[S7] OpenAI, Powering Product Discovery in ChatGPT (24 Mar 2026). Source — ACP expansion from checkout into product discovery.
[S8] Stripe, How to prepare for agentic commerce: technical field guide (10 Mar 2026). Source — Production integration lessons across merchants.
[S9] Mastercard + Worldline + ING, live European agentic payment (2 Jun 2026). Source — Live production payment evidence.
[S10] Visa + Artemis, Agentic Payments: What Onchain Data Reveals (14 Jul 2026). Source — Live onchain agent-payment evidence.
[S11] Mastercard, Verifiable Intent framework explainer (2026). Source — Cryptographic intent, identity and privacy model.
[S12] Reuters, Visa/Mastercard/Ant International Know-Your-Agent initiative (10 Sep 2026). Source — Latest interoperability initiative; framework under development.
[S13] Visa Trust Index for agentic commerce (9 Sep 2026). Source — Harris Poll; 2,065 US consumers.
[S14] Stripe, 10 lessons from first-generation agentic commerce (12 Mar 2026). Source — Production edge cases and merchant integration evidence.
[S15] Mastercard, Agent Pay for Machines (10 Jun 2026). Source — Machine-to-machine permissioning and multi-rail settlement.
[S16] IMF, How Agentic AI Will Reshape Payments (Apr 2026). Source — Independent policy/financial-system analysis of agentic payments.