What "Agent-Ready" Actually Means for Retailers Right Now

On 11 January 2026, Google announced the Universal Commerce Protocol at the National Retail Federation's Big Show, co-developed with Shopify, Etsy, Wayfair, Target and Walmart, and endorsed at launch by more than 20 organisations including Visa, Mastercard and Stripe. By April, the protocol's governing Tech Council had doubled in size, adding Amazon, Meta, Microsoft, Salesforce and Stripe. A separate protocol, the Agentic Commerce Protocol built by OpenAI and Stripe, powered ChatGPT's Instant Checkout feature from its September 2025 launch until OpenAI pulled back from the initial checkout model in March 2026, saying it wanted to give merchants more flexibility while shifting its focus toward product discovery. Agentic commerce moved from concept to competing, still-evolving infrastructure standards inside a single year.

What the market estimates actually say, and why they disagree

Morgan Stanley's own research puts US agentic commerce spending at $190 billion to $385 billion by 2030, representing 10% to 20% of online retail. McKinsey, in research published the same season, puts orchestrated US retail revenue at up to $1 trillion by 2030, with a global figure of $3 trillion to $5 trillion. Those numbers look wildly inconsistent sitting next to each other, and they are, but not because one firm is right and the other wrong. Both firms are using different scopes and definitions of what counts as agentic commerce, and neither publishes a methodology detailed enough for this piece to say precisely where the two definitions part ways.

That definitional difference matters more than either headline number. A retailer trying to size the opportunity, or the risk, needs to know which version of "agentic commerce" a given forecast is actually describing before deciding what it implies for their own roadmap.

What "agent-ready" actually requires

Strip away the market-sizing debate and UCP's actual technical scope is fairly specific. The protocol defines a set of discrete Capabilities, including Catalog for product discovery, Checkout, Identity Linking, Order and Payment Token Exchange, with Fulfilment handled as an extension. A business declares which of these it supports, and a platform, an AI agent, an app, or another system, discovers that support and invokes the relevant capability over REST, JSON-RPC, MCP or A2A. That's a genuinely broad architecture, and it's worth being precise that UCP itself doesn't define "agent readiness" as a single deliverable to tick off. What follows is our own reading of what operational readiness requires beyond the protocol's technical scope.

UCP's own documentation draws a distinction worth building on here: catalog responses covering pricing and availability are described as reflecting the business's current terms for that specific request, but they are explicitly not transactional commitments, and the documentation itself says such responses should be revalidated rather than reused across sessions, with the checkout process treated as the authoritative source for final terms. That's a sensible design choice: it builds in a mechanism for the retailer's terms to be re-checked before a transaction confirms. It also means the work of keeping the underlying commerce data accurate between one revalidation and the next still sits with the retailer rather than the protocol.

Why the protocol layer isn't the hard part

Implementing protocol support is an integration task. Keeping the commerce information behind it accurate, in real time, against a store's actual inventory, pricing, promotions and fulfilment rules, is a separate, ongoing operational commitment, and it's where the actual difficulty sits.

Consider a hypothetical retailer operating toward the more autonomous end of that spectrum, where an agent is authorised to complete a purchase without a final human confirmation step. If that retailer's product feed marks an item as in stock because the warehouse management system hasn't synced in the last few minutes, while the item actually sold out during a flash promotion, an agent transacting on that stale information completes a purchase the retailer can't fulfil. A human shopper hitting the same stale page might retry, call the store, or simply give up.

Today's most visible implementations don't typically work this way. OpenAI's Agentic Commerce Protocol, for instance, had the platform pass order details to the merchant's own backend, which could accept or decline the order before it was confirmed, and the checkout experience OpenAI built around it required the user to explicitly confirm each step. That's a meaningful safeguard, and it's part of why a stale feed's consequences are currently contained more often than not. But the safeguard lives in how much autonomy a given implementation chooses to grant rather than in the protocol itself, and any retailer or platform opting for less human confirmation takes on more of that risk directly.

The same problem shows up with policy as much as inventory. A promotion might only apply to loyalty members, or a return window might differ by product category, or a discount code might be restricted to first-time customers. None of that is necessarily encoded anywhere an agent can read it directly. If it isn't, an agent has no way to know the rule exists, and either applies a discount that shouldn't apply or fails to honour one that should.

This is a context problem wearing a protocol costume

None of this is a criticism of UCP or any other commerce protocol. The protocols solve the problem they're built to solve: giving agents a standard way to talk to any merchant instead of requiring a custom integration for each one. What they don't solve, and don't claim to, is whether the specific information an agent retrieves through that standard interface actually reflects the retailer's real-time inventory, current promotional rules and fulfilment constraints at the moment a transaction happens.

McKinsey's own 2026 retail research makes a related point from an independent angle. The firm's guidance for retailers describes product data, pricing logic, availability signals and fulfilment reliability as inputs that automated decision-making depends on, and its advice for becoming "agent ready" is explicit that this information needs to be structured and accurate in real time, rather than simply present somewhere in the business's systems. That's a useful, independently sourced confirmation that accuracy sits alongside protocol adoption as a separate requirement, even though McKinsey's research is addressing how agents select and transact rather than making a broader claim about organisational context.

That's the same distinction this publication has drawn in other contexts: having a live, technically correct data feed is not the same as having a system that reflects the actual, current state of the business behind it. A UCP integration that's technically compliant and slightly stale is a functioning protocol connection attached to an inaccurate picture of the store. Both halves matter, and protocol adoption only guarantees the first one.

What this means for retail leaders evaluating agent readiness

Two separate questions are worth asking before treating "we support UCP" or "we're integrated with an AI shopping surface" as a finished project. The first is the protocol question: does the business correctly declare and implement the capabilities it wants agents to use, covering the transaction types it actually wants to support. The second is the accuracy question: does the data behind those capabilities reflect inventory, pricing, promotional eligibility and fulfilment rules that are actually current at the moment an agent reads them, rather than just at the moment the integration was built.

A retailer that answers the first question well and the second poorly hasn't achieved agent readiness. It's built a fast, standards-compliant way for agents to act on information that might already be wrong.

AinTools works with retailers turning live commerce data into transactions that reflect their actual inventory, policies and fulfilment constraints. Our industries page sets out how we approach it.

Sources

• Universal Commerce Protocol, official specification and documentation: github.com/Universal-Commerce-Protocol/ucp
• Morgan Stanley Research, "Agentic Commerce Impact Could Reach $385 Billion by 2030": morganstanley.com
• McKinsey & Company, "AI-powered e-commerce: boost sales with agentic shopping": mckinsey.com
• CNBC, "OpenAI revamps shopping experience in ChatGPT after struggling with Instant Checkout offering" (24 March 2026): cnbc.com
• OpenAI Developers, Agentic Commerce documentation (merchant accept/decline flow, product feeds): developers.openai.com