AI & ECOMMERCE

The Dawn of Agentic Shopping

Artificial Intelligence in e-commerce was bound to happen. Yesterday, OpenAI announced Instant Checkout, a new feature that allows you to make purchases directly within a ChatGPT conversation.

No more redirects, no extra steps: just chat, choose, and buy.

But what’s particularly striking is who OpenAI partnered with. Not Amazon, not Temu, but Etsy and Shopify. That matters. Etsy is the marketplace for unique, vintage, and creative products. Shopify is the platform that enables anyone, from a small artisan to a global bran, to build their own online store.

This move may look small at first sight, but it opens an entire world of implications. Let’s break them down.

Agentic Shopping, Protocol Wars, and What Changes Next

OpenAI’s Instant Checkout turns ChatGPT into a point-of-sale: you can discover, decide, and buy inside a conversation.

The initial integrations (Etsy and Shopify) are not accidental; they sidestep platform giants that won’t surrender the customer relationship.

Under the hood, a new protocol layer (e.g., Google’s A2A/A2P and OpenAI+Stripe’s ACP) is emerging to standardize how agents talk to catalog, payments, logistics, and support systems.

This is not “voice shopping v2.” It’s a new interface for commerce (the conversational agent) plus a new infrastructure to make it safe, auditable, and scalable.

Below, I unpack the strategic logic, business upside, technical plumbing, operational risks, compliance angles, and practical steps for merchants, PSPs, banks, and marketplaces.

Why this matters (and why now)

• A new buying surface. For 25 years, checkout lived in websites and apps. Then social commerce embedded “buy” into feeds. Agentic commerce embeds “buy” into dialogue: ask, compare, tailor, finance, and confirm in one thread.

  
  

• Fewer clicks → higher intent. Chat collapses the funnel: question → clarification → recommendation → payment. Expect shorter time-to-decision, better conversion on mid-consideration products, and stronger attachment of services (warranties, financing).

  
  

• SME-first go-to-market. Amazon/Temu/Booking won’t let a third-party agent own checkout in their house. Etsy/Shopify aggregate millions of SMEs hungry for reach. This yields quick supply breadth without political wars with hyperscalers.

  
  

Why Etsy & Shopify (and not Amazon) is the right first move

• Control of the customer. Marketplace giants optimize for their interface. Allowing an external agent to execute checkout means ceding the most valuable asset: behavioral data + upsell moment.

  
  

• Catalog differentiation. Etsy offers unique/long-tail inventory that shines in advisory conversations (“Find me a handmade leather notebook under €50”). Shopify brings millions of stores across verticals perfect terrain to test agentic patterns.

  
  

• Partnership leverage. SMEs and enabling platforms are more open to experimental rails (protocols, new payout flows, agent APIs) than big-box incumbents.

  
  

• Opening the door to SMEs. By starting with Etsy and Shopify, OpenAI is clearly signaling a strategy: make agentic commerce directly accessible to small and medium merchants. Instead of depending on giants who will never let go of their customer flow, OpenAI is empowering the “long tail” of e-commerce to compete on equal footing. This positions ChatGPT as an alternative channel that levels the playing field against Amazon-scale ecosystems.

  
  

Open question: Is SME inventory sufficient for mainstream adoption? For commodity goods, the absence of Amazon-grade breadth may limit early usage. Expect hybrid approaches: agent suggests, then deep-links to large marketplaces when users insist.

The protocol layer: A2A, A2P, ACP (why you should care)

Think of these as TCP/IP for agentic commerce the rules for secure, auditable coordination between agents and enterprise systems.

• A2A (Agent-to-Agent, Google). Lets agents cooperate (e.g., a shopping agent talks to a returns agent). Expect standardized delegation, capability discovery, and contracts between agents.

  
  

• A2P (Agent-to-Pay, Google). Encapsulates payment intents, authentication, and receipts in an agent-safe way. Think: SCA prompts, tokenization, and dispute hooks built into the dialogue.

  
  

• ACP (Agentic Commerce Protocol, OpenAI + Stripe). Focused on cart, checkout, settlement. Standard objects for product, pricing, taxes, shipping, refunds, and disputesas API surfaces that an agent can safely call.

  
  

Why it’s strategic: If protocols become de facto standards, they determine who sets the rules on identity, trust, fees, and data portability. That’s why payments players, PSPs, and banks should engage now, not later.

What the user experience will look like (and where it wins)

A typical flow:

1. User: “I need a minimalist desk lamp under €80, warm light, fast shipping.”

2. Agent: Asks 2 clarifiers, composes a shortlist from connected catalogs.

3. User: Picks one; asks for a price match or an alternative with better reviews.

4. Agent: Applies coupon, checks stock, proposes delivery windows.

5. User: “Ok. Buy it.” → Instant Checkout triggers A2P/ACP flow with saved credentials, shows SCA prompt if needed, confirms order, sends receipt.

Where it wins:

• Advisory purchases (furniture, craft, gifts, gadgets): the dialogue adds value.

• Bundles & add-ons (warranty, accessories): better attachment in context.

• Services (classes, consultations, repairs): scheduling and payment in one.

• 
  

Where it struggles (initially):

• High-visual PDP needs (fashion fit & finish) unless the agent shows rich previews.

• Heavy comparison shopping where users enjoy browsing themselves.

• Complex cross-border logistics until agents master customs, taxes, returns.

  
  

Data, identity, and trust: the crux

• Who is the shopper? Agents will rely on verified profiles, stored payment tokens, and signed consents. Expect reusable “commerce identity” objects that travel across merchants under user control.

  
  

• Bias & disclosure. If an agent earns affiliate fees, users must see why it recommended option A. Disclosure and explainability (e.g., “lowest price + 2-day delivery + 4.8★ reviews”) will be table stakes.

  
  

• Privacy budget. Agents should reveal minimum necessary data to merchants (e.g., address only after shipping choice). Protocols will codify data minimization and auditable logs.

  
  

Discovery economics will change (again)

• SEO/SEM displacement. If the agent answers directly, fewer people hit a SERP or a category page. The gatekeeper shifts from search engines/social feeds to assistant surfaces.

  
  

• From ads to bounties. Expect agent-safe offers: dynamic coupons, CPA bounties, and preference contracts. Merchants will bid not just for eyeballs but for agent attention with clear constraints (no dark patterns).

  
  

• Brand storytelling still matters. Agents can summarize reviews, but brand trust and post-purchase experience will still drive retention especially for higher-ticket items.

Returns, fraud, and support: where human-in-the-loop stays crucial

• Returns orchestration. The agent can generate labels, schedule pickups, and track refunds if the merchant exposes clean APIs. But edge cases (damaged goods, partial refunds) still need human review.

  
  

• Conversation-native fraud. New vectors appear: synthetic identities guiding agents to exploit promo rules; prompt-based item swaps; image edits to claim defects. Merchants must add agent-aware fraud signals (conversation telemetry, device binding).

  
  

• Escalation design. Agents should offer human transfer at sensible thresholds (value, risk, sentiment). “Human at the right time” will be a competitive differentiator.

  
  

What merchants must change (operations & tech)

Operational realities:

• Structured catalogs. Agents need clean product graphs (attributes, rich media, variants, availability). Messy data → bad recommendations.

• Live inventory & shipping SLAs. Dialogues promise immediacy. If “in stock” lies, trust collapses. Wire real-time stock and carrier ETA (estimated time of arrival) feeds.

• Policy codification. Returns, warranties, and price-match rules must be machine-readable so the agent can negotiate within guardrails.

Tech checklist:

• Expose ACP/A2P adapters (cart, totals, taxes, pay, refund).

• Provide offer/intents APIs (discounts, bundles, loyalty redemption).

• Implement receipts & dispute endpoints that agents can trigger.

• Add telemetry: track agent-origin sales, conversion, AOV, refund rate.

KPIs to watch:

• Agent conversion vs. web/app, time-to-purchase, AOV (average order value) uplift, attach rates (warranty/financing), refund/chargeback deltas, LTV of agent-acquired customers.

  
  

Payments & financing: embedded where the decision happens

• SCA and compliance in-chat. Expect secure prompts, biometric fallbacks, and signed proof of intent baked into the protocol.

  
  

• Instrument orchestration. Cards, wallets, account-to-account (A2A payments), and BNPL offered contextually (“Split in 3 at 0%?”).

  
  

• Risk and disputes. Conversation logs become evidence: agent explained fees, delivery ETA, and refund policy. PSPs/banks should adapt dispute tooling to accept agent artifacts.

  
  

• Settlement complexity. Multi-merchant carts will need smart split settlement and tax handling. Protocols should standardize multi-party payouts.

  
  

Regulation: consumer protection meets AI alignment

• Consumer law. Clear consent, price transparency, cooling-off periods, and fair refund rules must be enforced by design. Agents should refuse non-compliant flows.

  
  

• Payments law. PSD2/PSD3 (EU), Reg E (US), strong authentication, data protection, agent UIs must be compliant without dumping users back into web views.

  
  

• Platform law. If assistants rank offers, expect DMA/DSA-style scrutiny (self-preferencing, explainability, ad disclosure). Keep ranking signals auditable.

  
  

Accessibility, inclusivity, and languages

Conversational commerce can be more inclusive: voice, low-vision support, plain-language explanations, automatic translation. If you sell cross-border, agents can bridge language gaps and access needs seamlessly if your data and policies are machine-readable.

Competitive responses

• Amazon/Temu/Booking. Likely to deepen their own agents inside their walled gardens, tighten loyalty hooks, and restrict external agent checkout. Also: push their private protocols and require merchants to adopt them.

  
  

• Apple & Google. Own the dominant user surfaces (iOS/Android). Expect native assistant commerce tightly integrated with OS identity and payments (Pay/Wallet), and in Apple’s case, a premium privacy story.

  
  

• PSPs/Banks. The smart move is to ship agent-native payment APIs, dispute endpoints, and financing modules that speak A2P/ACP. Become the trusted financial layer for agents.

  
  

A practical playbook (what to expect next)

For merchants/brands

• Clean your catalog data (attributes/images/reviews/specs).

• Publish policy as code (returns, warranty, price match).

• Wire real-time inventory & shipping.

• Pilot an ACP/A2P adapter with a limited SKU set.

• Design escalation to human and track agent-origin metrics.

For PSPs/banks

• Ship agent-safe payment/identity APIs (SCA, tokens, receipts).

• Accept conversation logs as dispute evidence; update ops playbooks.

• Offer embedded financing and loyalty hooks as protocol extensions.

For marketplaces/platforms

• Decide your stance: open discovery vs closed checkout.

• Provide quota-controlled discovery APIs (search, price, stock).

• Build brand guardrails: how agents use your ratings, reviews, and content.

For regulators

• Codify disclosure, consent, ranking explainability.

• Certify protocol compliance (consumer rights baked into flows).

• Encourage interoperability to avoid agent monopolies.

Closing thought

“Instant Checkout” looks like a small UI tweak. It isn’t. It’s the opening move in a shift where the conversation becomes the store, and protocols become the rails. The winners won’t simply be those with the biggest catalogs or the lowest fees, but those who can speak the language of agents, prove trust & compliance by design, and blend automation with timely human judgment.

If your roadmap still assumes customers will start at your homepage, scroll your PDPs, and tap your native checkout you’re planning for yesterday. Start designing for the moment the purchase begins with a question.

-Nacho Beites-