In the first three weeks of February 2026, OpenAI shipped Instant Checkout inside ChatGPT, Amazon made Alexa+ free for 250 million Prime members, Meta committed up to $135 billion toward agentic commerce infrastructure, and Microsoft declared that Copilot would become “the new front door to retail.” Then the consumer data arrived: a PYMNTS survey of 2,299 U.S. adults found that while 70% express interest in AI-powered shopping, only 17% feel comfortable actually completing a purchase through an agent.
The gap between launch activity and consumer readiness isn’t a marketing problem. It’s an infrastructure problem. Every platform shipped the ability to find products, compare prices, and initiate transactions. None shipped the trust architecture that consumers are explicitly demanding: hard spending limits, credential security, human approval before money moves, and a clear record of what the agent bought and why. Bain & Company reports that 95% of consumers still perform at least one manual verification step before letting an agent complete a purchase. The checkout feature works. The checkout trust doesn’t exist yet.
This is exactly the problem AgentPMT was built to solve. While the checkout wars focus on who controls the transaction, we built the trust layer that makes transactions safe regardless of which platform hosts the conversation. AgentPMT’s Agent Credit Card Integration keeps payment credentials completely out of the agent’s context — credentials are injected server-side at the moment of purchase and never appear in logs, prompts, or model memory. Combined with budget controls that enforce hard daily, weekly, and per-transaction spending limits, and Human-in-the-Loop flows where agents pause and request approval via mobile app before completing a purchase, AgentPMT addresses each of the specific trust gaps the consumer data identifies. The checkout wars aren’t about who builds the best shopping agent. They’re about who builds the infrastructure that makes consumers willing to hand over their wallets.
The February Checkout Blitz
The pace of launches was staggering. On January 28, Mark Zuckerberg told investors that Meta’s agentic commerce tools would ship “within months,” describing a vision of “personal superintelligence” powered by social graph data across Facebook, Instagram, and WhatsApp. Meta’s $115-135 billion 2026 capital expenditure budget and its acquisition of AI agent startup Manus signaled how seriously the company is treating this bet.
A week later, Amazon made Alexa+ available to all U.S. users, free for Prime members. The dual-AI architecture — Amazon Nova for speed, Anthropic Claude for reasoning — is already producing results: users tripled their shopping activity, and conversations increased 2-3x compared to the original Alexa. Amazon integrated Ticketmaster, Uber, OpenTable, and a dozen other services, positioning Alexa as a general-purpose commerce agent. Meanwhile, approximately 50 million daily shopping queries now flow through ChatGPT across more than 800 million weekly active users, per Modern Retail.
Then came the defining launch. On February 16, OpenAI unveiled “Buy it in ChatGPT” — Instant Checkout powered by the Agentic Commerce Protocol, co-developed with Stripe. The protocol is open source under Apache 2.0, and existing Stripe merchants can enable agentic payments with as little as one line of code. Etsy sellers went live immediately. Over one million Shopify merchants — including Glossier, SKIMS, Spanx, and Vuori — are next. PayPal will bring tens of millions more. As Stripe’s Will Gaybrick stated, the company is “building the economic infrastructure for AI” and “re-architecting today’s commerce systems.”
Microsoft positioned Copilot Checkout as a “Personalized Shopping Agent” on February 9, announcing a Mastercard Agent Pay partnership. Uber launched Cart Assistant for grocery shopping on Uber Eats, working with Albertsons, Kroger, Safeway, and Sprouts. This is more consumer-facing AI commerce infrastructure in three weeks than the previous two years combined.
Amazon CEO Andy Jassy offered a counter-narrative worth weighing. Speaking to shareholders on February 10, Jassy argued that “a lot of customers are ultimately going to choose to use a great shopping agent from that retailer” over horizontal third-party agents. His reasoning: horizontal agents “don’t have any of your shopping history” and “get a lot of the product details wrong.” Amazon’s own Rufus agent was used by 300 million customers in 2025, and Rufus users were 60% more likely to complete a purchase — numbers that support the first-party advantage thesis. But Jassy’s argument also reinforces the fragmentation problem: if every retailer builds its own agent, businesses deploying purchasing agents need infrastructure that works across all of them. AgentPMT’s Dynamic MCP connects agents to tools and services across platforms with one integration, so builders don’t have to rebuild trust architecture for every shopping channel.
The 70-17 Problem
The consumer data tells a remarkably consistent story across multiple independent studies. PYMNTS found 70% of consumers express interest in AI agents for subscription management, grocery shopping, and meal planning — but only 17% feel comfortable completing a full purchase. XCCommerce, in a study with the National Retail Federation, found that while 70% of shoppers use AI to seek promotions, comfort levels break down sharply by task: 54% use AI to find the best price, 42% for comparing specifications, 41% for summarizing reviews. Only 4% would let an AI complete a purchase outright.
The pattern is unmistakable. Consumers enthusiastically use agents for research, discovery, and comparison. They hit a hard wall at the moment money changes hands.
The invidis analysis quantified this as the “verification tax” — AI saves research time, but consumers spend equal time fact-checking the output. Their data shows 58% of shoppers use AI to research products, but the conversion rate for AI-assisted purchases sits at roughly 2%. Bain & Company adds another dimension: 60% of searches are now “zero-click,” meaning AI provides summaries without consumers visiting retailer sites. The agent is already mediating discovery. It just can’t close the sale.
The consumer concerns are specific and solvable. They worry about agents spending too much. They worry about agents accessing payment credentials. They worry about purchasing the wrong item with no way to verify before it’s final. Among consumers who don’t regularly use AI tools, trust in generative AI platforms as purchasing agents sits at 3%, per PYMNTS. Banks and digital wallets rank as the most trusted entities for agent-mediated commerce — not AI platforms.
This is a specification, not a rejection. Consumers are saying: give me control before the money moves. AgentPMT’s Human-in-the-Loop was designed for exactly this interaction pattern. The agent works through research, comparison, and selection autonomously, then sends an approval request to your phone before completing the purchase. You review. You approve or reject. The agent proceeds. This is the co-pilot model that 95% of consumers are demanding through their behavior — not autopilot, but a collaborator that knows when to ask permission.
The advertising disruption compounds the stakes. As Retail Brew reported, brands may soon offer ChatGPT time-sensitive discounts during agent searches — a $20-off coupon surfaced only while the agent evaluates options. “A lot of the ads in the future with agentic shopping will really be behind the scenes,” noted Andy Szanger of CWD. When the customer becomes an algorithm, the entire digital advertising industry faces a structural reckoning.
When Agents Shop Where They’re Not Welcome
The consent problem is the checkout wars’ first structural failure. CNBC reported that Amazon’s “Buy for Me” feature surfaced products from third-party retailers who never opted in. The feature grew from 65,000 products to more than 500,000 within months. Retailers discovered they were part of the program when orders arrived from a “buyforme.amazon” email address. One Virginia-based stationery shop owner told CNBC the experience plainly: “Sounds like a great program until the agentic AI starts selling customers things you don’t have.”
Amazon directs retailers to email an opt-out address to be removed — placing the burden on the merchant, not the platform. Meanwhile, Amazon has blocked dozens of AI agents from accessing its own site and sued Perplexity for allegedly concealing agents that scraped Amazon’s listings.
Then there’s the fee structure. Shopify merchants using ChatGPT checkout pay OpenAI 4% on top of existing Shopify fees — a combined take rate of approximately 9.2%. Google and Microsoft are charging no additional checkout fees, at least for now. Shopify’s Harley Finkelstein reported that AI-driven traffic to Shopify stores increased 7x and AI-attributed orders grew 11x over the past year. The traffic is real. The question is who captures the margin.
The fragmentation compounds the cost problem. Merchants now face three competing standards: ACP from OpenAI and Stripe, UCP from Google with Shopify, Target, and Walmart, and proprietary systems from Amazon and Meta. Each requires separate integration, separate fee structures, and separate data-sharing agreements. This is the early mobile app development problem again — build for every platform or miss revenue.
AgentPMT’s approach inverts this dynamic. Rather than letting platforms dictate which vendors the agent can transact with and at what fee, AgentPMT puts the business owner in control. Vendor whitelisting lets you approve exactly which merchants your agents can transact with. Product-level restrictions control which tools each budget can access. And complete cost transparency — per-tool pricing visible before and after every call — means no hidden take rates. The checkout wars are creating new gatekeepers. The alternative is infrastructure that puts the buyer, not the platform, in control.
What the Trust Stack Actually Requires
The consumer data maps to five specific trust requirements: credential security, spending controls, verification before purchase, audit trails, and vendor transparency. Every checkout platform launched in February solved part of this stack. None solved all of it.
Stripe’s Shared Payment Token addresses credential security at the protocol level — payment details are encrypted and authorized only for specific amounts and specific merchants. That’s a genuine advance. But SPT doesn’t impose spending caps, doesn’t pause for human approval, and doesn’t generate governance-level audit trails. Amazon Alexa+ brings deep shopping history and personalization but locks users into a proprietary ecosystem. Google’s Universal Commerce Protocol is an open standard, but it relies on merchants adding a /.well-known/ucp endpoint and provides no governance layer. Meta brings the strongest personal context — social graph data across billions of users — but carries the weakest trust signal, because consumers historically view social media platforms as poor stewards of purchasing data.
Uber’s Cart Assistant offers a positive signal for how trust gets built incrementally. The tool learns users’ preferences over time, showing favorite brands and accounting for store-level availability. As Uber CTO Praveen Neppalli Naga described it, the approach starts “with real customer needs and building practical solutions.” Trust through familiarity, not through transaction features. But it still lacks the governance architecture for when agents start spending autonomously.
AgentPMT was designed as the complete trust stack, platform-independent and not locked to any single checkout standard. Agent Credit Card Integration: credentials never enter the agent’s context, never appear in logs, never accessible to the model. Budget controls: hard daily, weekly, monthly, and per-transaction limits enforced server-side. Human-in-the-Loop: agents send approval requests to your phone and pause until you respond. Auditable Everything: full request/response capture, workflow step tracking, compliance-ready audit trails. Vendor whitelisting: you define where your agent shops. Whether your agent runs on ChatGPT, Alexa, Copilot, or a local model, the trust infrastructure works because it operates at the governance layer, not the platform layer.
What This Means for You
If you’re a merchant, the checkout wars force an immediate decision: which AI shopping platforms do you integrate with? The fee structures are already diverging — 4% on ChatGPT versus 0% on Google and Microsoft — and the consent models vary from opt-in to involuntary inclusion. Evaluate each platform on fee structure, data ownership, and whether your products appear by default or by choice.
If you’re deploying purchasing agents for your business or customers, the consumer data provides a clear implementation order: build the approval flow first, the autonomy second. Spending controls, credential isolation, and audit trails aren’t premium features — they’re prerequisites for adoption. AgentPMT provides that governance layer across every platform. The trust stack — credential security, budget enforcement, human verification, audit trails, vendor controls — works whether agents purchase through ACP, UCP, or proprietary channels, because the trust requirements are universal.
What to Watch
Monitor OpenAI’s first shopping metrics, expected Q1-Q2 2026 — conversion data from 800 million weekly users will either validate the checkout model or confirm the trust gap at scale. Track Shopify’s ACP adoption rate among one million merchants to gauge whether 4% fees are acceptable. Watch Amazon Alexa+ shopping revenue in Q1 earnings for attribution data behind the 3x activity increase. And follow the regulatory front: more than a dozen U.S. states have proposed laws targeting AI-driven “surveillance pricing,” and New York’s Algorithmic Pricing Disclosure Act could reshape how agents present prices entirely.
The most important number to track is the 17%. If that purchase comfort rate hasn’t moved by Q3 2026, the checkout wars will have produced five front doors to a store nobody walks into.
The checkout wars proved one thing: every major tech company believes agents will mediate trillions in commerce. McKinsey projects up to $1 trillion in U.S. retail revenue orchestrated by agents by 2030. They’re right about the destination. But the companies deploying shopping agents without trust infrastructure are building storefronts without locks on the doors. Ninety-five percent of consumers want control before they hand over their wallets. The winners won’t be the platforms with the most merchants or the lowest fees. They’ll be the ones that built the infrastructure to earn trust. AgentPMT provides that infrastructure — agent purchasing with credential isolation, hard budget enforcement, human approval flows, and complete audit trails — across every platform, for any agent.
Key Takeaways
- Five major tech platforms launched consumer AI shopping agents in February 2026, but only 17% of consumers are comfortable completing a purchase through an agent — the trust gap, not the feature gap, determines who wins
- Consumer requirements are specific and solvable: credential security, hard spending limits, human verification before purchase, complete audit trails, and vendor transparency — every platform solved part of this stack, none solved all of it
- Merchants face fee fragmentation (4% ChatGPT vs 0% Google/Microsoft) and consent battles (Amazon listing products without permission) — evaluating platform integration requires weighing margin, data ownership, and control
Sources
- Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol — OpenAI Blog
- Shopify Merchants to Pay 4% Fee on ChatGPT Checkout Sales — PYMNTS
- Alexa+, Amazon’s AI assistant, is now available to everyone in the US — TechCrunch
- Zuckerberg teases agentic commerce tools and major AI rollout in 2026 — TechCrunch
- Amazon CEO says shoppers will choose retailers over third-party AI platforms — Retail Brew
- 70% of Consumers Say Yes to AI Agents for Shopping — PYMNTS
- Study Reveals 70% of Shoppers Use AI to Seek Promotions — Retail Insider
- Developing an open standard for agentic commerce — Stripe Blog
- How agentic commerce is becoming the new front door to retail — Microsoft Industry Blog
- What will ads in an agent-assisted shopping world look like? — Retail Brew
- Amazon’s AI agents spark backlash from retailers after listing products without permission — CNBC
- Agentic Commerce: Why AI is Winning the Search War but Losing the Checkout Battle — invidis
- Online Shopping Could Be AI’s Next Victim — Bloomberg
- Why the AI shopping agent wars will heat up in 2026 — Modern Retail
- What Happens to Stores When AI Agents Do the Shopping? — PYMNTS
- Uber debuts AI-powered personal shopper feature — Grocery Dive
- States Push New Laws Targeting AI ‘Surveillance Pricing’ — Skift