March 30, 2026

Shoptalk Made It Official: Agentic Commerce Is the New Default

March 24-30, 2026 - Shoptalk turned agentic commerce from industry chatter into corporate strategy. Here’s what happened.

Last week we covered the stack assembling in real time - Perplexity’s injunction reversed, Shopify defaulting all stores into ChatGPT, Visa onboarding 21 European banks. This week, the industry gathered at Shoptalk in Las Vegas, and every major announcement confirmed the same direction.

The headline number: Bain & Company projected US agentic commerce could reach $300 to $500 billion by 2030. But the real story isn’t the projection. It’s that the biggest retailers stopped talking about “if” and started announcing “how.”

Walmart and Sephora launch branded apps inside ChatGPT

During Shoptalk week, Walmart debuted Sparky as a dedicated app inside ChatGPT. We covered last week that Sparky was going multi-platform. Now it’s live. Customers can link their Walmart account, access loyalty programs, and check out through Walmart’s own payment system - all without leaving the ChatGPT conversation.

The same day, Sephora launched its own ChatGPT app. Users prompt beauty advice (“Sephora, help me find a foundation for dry skin”), link their Beauty Insider loyalty account, and get curated recommendations. Checkout integration is planned but not live yet - Sephora is starting with discovery and personalization, adding transactions later.

Both announcements validate the model we described when OpenAI killed Instant Checkout: AI handles the conversation, retailers own the checkout. But Walmart added a revealing data point. Its EVP told Search Engine Land that ChatGPT’s Instant Checkout converted 3x worse than when users clicked through to Walmart’s own website. That’s the number that killed in-chat checkout - and that’s why every retailer is now building branded apps instead.

Why it matters for merchants: The retailer-app-inside-ChatGPT model is now the confirmed path for large retailers. Walmart, Sephora, Target, Instacart, DoorDash, Expedia - the list keeps growing. Small and mid-size merchants won’t build ChatGPT apps. Their path remains product data quality through Shopify’s default integration or direct ACP feeds. But the 3x conversion gap should reassure merchants worried about losing checkout control to AI platforms: users still prefer buying from brands they know.

Gap goes all-in on Google Gemini checkout

The biggest Shoptalk surprise came from fashion. Gap announced that all its brands - Gap, Old Navy, Banana Republic, Athleta - will enable full checkout inside Google Gemini through Google’s Universal Commerce Protocol (UCP). Checkout via Google Pay, shipping handled by Gap.

But the real innovation is the Agentic Sizing Protocol (ASP), developed with Bold Metrics. It gives AI agents access to fit data, so when a customer asks Gemini for jeans that fit like their favorite pair, the agent can actually recommend the right size across Gap’s brands. If the sizing problem gets solved, it removes one of the biggest barriers to fashion purchases through AI.

Retailer	AI platform	Protocol	Status	Checkout
Walmart	ChatGPT	ACP	Live (March 24)	In-app (Sparky)
Sephora	ChatGPT	ACP	Live (March 24)	Coming soon
Gap (all brands)	Google Gemini	UCP	Announced	Google Pay
Target	ChatGPT + Gemini	ACP + UCP	Rolling out	In-app
Best Buy	ChatGPT	ACP	Live	On-site

Why it matters for merchants: Gap choosing Google Gemini (UCP) over ChatGPT (ACP) as its primary AI commerce channel is significant. Fashion brands have a specific data problem - sizing - and UCP’s open protocol structure lets them add custom capabilities like ASP. For fashion and apparel merchants: structured size and fit data is now a competitive advantage in AI discovery, not just an e-commerce best practice.

Stripe and Tempo launch Machine Payments Protocol

Just before Shoptalk, on March 18, Stripe and payments network Tempo launched the Machine Payments Protocol (MPP), an open standard for AI agents to make payments programmatically. MPP supports stablecoins, traditional card payments, and Bitcoin Lightning.

The same day, Visa extended MPP with a card specification and SDK, connecting the protocol to Visa’s global network. This means AI agents using MPP can process payments on existing card rails - not just crypto.

Meanwhile, BigCommerce integrated Stripe’s Agentic Commerce Suite, giving its merchants agentic discovery and checkout with fraud protection via Shared Payment Tokens and Stripe Radar.

Why it matters for merchants: The payments stack for agentic commerce is getting concrete. Stripe’s SPTs (which we’ve covered since early February) now have an open protocol underneath them. For merchants on BigCommerce, agentic commerce payments are now a platform feature, not a custom integration. The stack is: discovery via ACP/UCP, payments via SPTs/MPP, fraud prevention via card network infrastructure. Merchants don’t need to build any of this - they need to make sure their product data feeds into it.

eBay joins Amazon in banning third-party AI agents

eBay updated its Terms of Service to explicitly ban third-party “buy-for-me” AI agents, LLM-driven bots, and any automated flow that places orders without human review. Amazon had already done the same - and is actively litigating it against Perplexity.

The paradox: both Amazon and eBay are building their own first-party AI shopping agents (Amazon’s Rufus and eBay’s own AI shopping tools, including an OpenAI Operator integration). The message is clear - AI agents are welcome on these platforms, but only if the platform controls them.

On Amazon specifically, Rufus now serves 250 million monthly active users, and Sponsored Products Prompts are now generally available. Only 22% of Rufus recommendations match traditional first-page results - meaning sellers’ existing Amazon SEO strategies are becoming less relevant.

Why it matters for merchants: The walled-garden vs. open-protocol split is hardening. Amazon and eBay are building closed AI commerce ecosystems. Shopify, Google, and the UCP coalition are building open ones. If you sell on Amazon, you need to optimize for Rufus specifically - and your traditional Amazon SEO playbook is increasingly outdated. If you sell on your own store, the open protocols (ACP, UCP) are your path to AI discovery.

Accenture bets real money on agentic commerce

Accenture Ventures invested in DaVinci Commerce, a company building “Agentic BrandStore” - branded conversational commerce experiences inside LLMs. Strategic partnership with Accenture Song for go-to-market. Both showcased at Shoptalk.

This is notable not for the technology (which is early) but for the signal. When a major consulting firm puts investment dollars behind agentic commerce infrastructure, it means enterprise clients are asking for it. The space is moving from “interesting experiment” to “line item in enterprise budgets.”

Regulation: the gaps are getting attention

The Center for Data Innovation published a report highlighting a concrete regulatory gap: Regulation E, the federal rule governing consumer dispute rights for electronic fund transfers, has no framework for agentic commerce. If an AI agent makes a purchase you didn’t intend, your dispute rights are unclear. NIST plans to host a public-private conversation on AI agent standards in April 2026.

In Europe, the EU AI Act omnibus amendments continue advancing. MEPs reached a preliminary political agreement on March 11, with committee votes on March 18. Most remaining AI Act rules take effect on August 2, 2026. How these rules apply to AI shopping agents - which process personal data, make purchasing decisions, and access merchant systems - is still being interpreted.

Why it matters for merchants: Consumer dispute rights for AI agent purchases are unresolved. If a customer’s AI agent buys the wrong product, who handles the return? The merchant, the AI platform, or the payment provider? Until regulation catches up, merchants should document their return and dispute policies for AI-initiated purchases clearly - even if the volume is small today.

The Shoptalk consensus

Stepping back from individual announcements, Shoptalk 2026 produced a clear consensus on where agentic commerce stands:

What’s working now: Discovery. Product research through AI. Comparison shopping. This is the proven use case, confirmed by Walmart’s 3x conversion data and OpenAI’s pivot.

What’s launching now: Retailer apps inside AI platforms. Branded checkout experiences. Protocol-based integrations (ACP for ChatGPT, UCP for Gemini and beyond).

What’s still missing: Universal carts (buying from multiple merchants in one AI conversation). Standardized product data across protocols. Clear liability frameworks for AI-initiated purchases. And, critically, any real traction beyond the largest retailers.

The eMarketer takeaway captured the tension: the strongest near-term impact of agentic commerce is in search, merchandising, and supply chain optimization - not end-to-end autonomous purchasing. The $300-500 billion projection is real, but the bulk of that value comes from AI improving how products are found and presented, not from agents buying autonomously.

What merchants should do this week

1. Audit your product data for AI discovery. Every Shoptalk announcement reinforced the same point: the merchants who win in agentic commerce are the ones whose product data is complete, structured, and machine-readable. If you’re on Shopify, check your AI feed now.

2. If you’re in fashion or apparel, add structured sizing data. Gap’s Agentic Sizing Protocol signals that fit data is becoming a competitive advantage. Structured size guides, fit descriptions, and measurement data help AI agents recommend your products with confidence.

3. Understand the platform split. Amazon and eBay are walled gardens with their own AI agents. Shopify, Google, and the UCP ecosystem are open. Your strategy should match: optimize for Rufus if you sell on Amazon, optimize for ACP/UCP if you sell on your own store. Ideally, do both.

4. Watch the Regulation E gap. Consumer dispute rights for AI purchases are unclear. Document your policies for returns and disputes on AI-initiated orders now, before volume grows and the edge cases multiply.