Agentic commerce is live, the funnel is collapsing into a single delegated decision, and most digital estates aren’t ready to be read by the machines doing the shopping

By Saurabh Sachdeva, Founder & CEO, Assurex

In my last piece I argued that AI visibility tactics get brands found, not chosen – that the recommendation is decided by a reasoning layer that reads the customer, not the page. I ended with a warning: all of that is before agentic commerce arrives.

It has arrived. Faster than almost anyone planned for.

This stopped being theoretical while most brands weren’t looking

A quick tour of what is now live, not roadmapped:

ChatGPT users in the US can buy products directly inside the chat – the assistant surfaces products, shows a buy button, and processes payment without the shopper ever visiting the merchant’s website. It launched on the back of the Agentic Commerce Protocol, an open standard co-developed by OpenAI and Stripe, first released in September 2025 and expanded to all US user tiers in early 2026. Etsy sellers went live first; over a million Shopify merchants are following.

Google answered in January 2026 with its own standard – the Universal Commerce Protocol – backed by a coalition including Walmart, Target, and Shopify, and wired into AI Mode in Search and Gemini. Microsoft’s Copilot Checkout is live in the US. The card networks are building the payment rails underneath: Visa and Mastercard have both launched agent-payment frameworks that let AI agents initiate and settle transactions using tokenised credentials, bounded by user-set spending limits.

Two competing commerce protocols, three major AI surfaces transacting, and both card networks racing to underwrite it. Analysts project this channel in the trillions of dollars by 2030. Whatever your view of the projections, the infrastructure argument is over. The question left is whether your brand can be read, compared, and bought by a machine – because that machine is now standing in your shop.

The customer in your funnel is increasingly not a person

Here’s what changes strategically, and it’s more profound than “a new checkout button.”

For twenty years, digital commerce has been an exercise in persuading a human moving through a funnel: awareness content, retargeting, landing pages, urgency banners, basket-recovery emails. Every discipline in the modern marketing stack assumes a human pair of eyes making a sequence of micro-decisions you can influence.

An agent shopping on a customer’s behalf compresses that entire sequence into one delegated decision – and it happens somewhere you can’t see. The agent reads the customer’s request, builds a shortlist from product data and third-party evidence, compares prices and terms across retailers, and executes. No session on your site. No impression to retarget. No basket to recover. The first signal you receive may be the order itself.

And the “shopper” that did visit – the agent – is far less susceptible to everything your conversion stack was built to do. It doesn’t feel urgency. It doesn’t respond to hero imagery. It reads structured substance: price, availability, specifications, delivery terms, returns policy, and what independent sources say about whether your product does what you claim.

This is my previous argument, hardened into a transaction. The reasoning layer used to decide what to recommend. Now it decides what to buy.

What “ready” actually means

The uncomfortable audit question for any commerce leader this year: if an AI agent arrived at your digital estate today with a customer’s mandate and a spending limit, could it complete its job? For most enterprise estates the honest answer is no – and the gap has three layers.

Layer one: can agents read you? Product data as machine-consumable feeds, accurate structured markup, real-time price and availability, explicit delivery and returns terms. This is the hygiene layer from my last piece – except the cost of failure has escalated from “misunderstood” to “silently skipped.” An agent that can’t parse your catalogue doesn’t ask for clarification; it buys from the competitor it can parse.

Layer two: can agents transact with you? Protocol readiness – and realistically, plural. The market has already split between the OpenAI/Stripe standard and Google’s coalition, and most brands will end up needing to support both, the way they once needed to be on both major app stores. For organisations on modern composable platforms, this is an integration project. For those on legacy monoliths with checkout logic welded to a human-only web journey, it’s an architecture conversation – and the honest lead time on those means the decision point is now, not when the channel matures.

Layer three: would an agent choose you? This is where everything from my previous piece compounds. The agent’s selection runs on value-fit and verifiable evidence: how your price compares in real time, what your delivery promise is against the customer’s need, what independent reviews and community sentiment say. A brand can be perfectly readable, perfectly transactable – and consistently passed over, because the machine arithmetic of its offer doesn’t win for the customer profile in question. No protocol integration fixes an uncompetitive offer. That remains a business strategy problem, now adjudicated at machine speed.

The trap to avoid: rebuilding the old playbook for a new buyer

Predictably, an industry is forming to sell “agent optimization” – and some of it will be the GEO story again: tactical tricks to game the shortlist. The same logic from my last piece applies, with higher stakes. Agents cross-check. Their platforms have every incentive to punish manipulation, because a user whose agent buys badly stops delegating. The durable play is not tricking the machine buyer; it’s being the offer the machine buyer correctly selects – and making every fact about that offer effortless to verify.

There’s also a genuine brand question that deserves board-level attention rather than a tactical response: when the agent owns the interaction, where does your customer relationship live? The brands that thrive in this channel will be the ones whose product is strong enough to be chosen sight-unseen, and whose post-purchase experience – fulfilment, support, quality – builds the kind of reputation that shows up in the third-party evidence agents read. The moat moves from the storefront to the operation behind it.

Where to start

Three moves I’d put in front of any enterprise commerce team:

Run the agent audit. Take a real product. Attempt to discover, evaluate, and (where protocols allow) purchase it through the major AI surfaces. Document where the journey breaks – unparseable data, missing feed fields, checkout walls an agent can’t cross. That gap list is your roadmap, and it’s usually sobering.

Sequence the protocol work. Assess what ACP and UCP support requires against your current architecture. Composable, API-first estates can move in weeks; monolithic ones need the honest conversation about what that means for platform strategy – which is a conversation worth having before the channel forces it.

Re-run your value proposition through machine eyes. Strip the brand storytelling and look at what an agent sees: price, terms, proof. If that stripped-down offer doesn’t win for a defined customer segment, that’s the finding that matters most – and no integration budget will paper over it.

The last two decades rewarded brands that mastered persuading humans on screens. The next one will reward brands whose offers survive being read by machines acting for those humans. The window where you can prepare calmly – before your category’s transactions visibly shift – is open now, and it won’t announce when it’s closing.

If you want to know what an agent sees when it reads your estate today, that’s an assessment we run. Start the conversation.

Saurabh Sachdeva is the Founder and CEO of Assurex, a digital engineering agency and Sitecore Platinum Partner with teams across London, Dubai, and India, helping enterprise brands build composable digital experiences. This is part two of a series; part one, Visibility Is Not Recommendation, argued that AI recommendation rewards customer fit over content optimization.