Summary: Refocusing on a Merchant Message as Merchants control these early days of Agentic
Something quietly changed in Google’s agentic commerce story over the last twelve months, and almost nobody has called it out.
As I related in Google I/O 2025 the headline was “Buy for me.” The pitch was clean and consumer facing: AI finds the product, you set the price you are willing to pay, Google watches the market, you confirm, and Google completes the checkout using Google Pay. Google called it agentic checkout. I wrote at the time that the price of entry for merchants was simply adding the GPay button, and that qualified agentic demand would finally give merchants a reason to do it.
Fast forward to 2026. Buy for me is barely mentioned. What Google talks about now is AI Mode, Gemini Shopping, Universal Commerce Protocol (UCP), Business Agent, and Agent Payments Protocol (AP2). The autonomous purchase capability still exists (just no longer the story). We have now moved to a retailer focused message as GC, Gemini and all the other Google assets enable every retailer expose inventory, offers, checkout, and post purchase service to agents.
The pivot from consumer feature to merchant platform, and I think it is the right call.
What the Silence Tells You – No Buy for Me Success Stories…
Google publishes merchant metrics when merchant metrics are good. That is what a platform business does. So look at what has not been published.
There is no case study claiming Buy for me cut cart abandonment by some impressive percentage. No return rate disclosure. No merchant testimonial about incremental conversion from autonomous checkout. Google has published plenty on AI shopping, conversational shopping, merchant integrations, Business Agent, UCP, and AP2. On the one feature that actually took the wheel away from the shopper, there is quiet.
Absence of evidence is not evidence of failure. But it is a signal about where Google is in proving business value, and it lines up with what the data says about consumer appetite. Bain’s work on autonomous shopping found roughly half of consumers are not comfortable letting AI handle a transaction end to end without them, and that consumers trust retailer owned agents about three times more than third party agents to actually complete the purchase ( Bain & Company, “Agentic AI in Retail: How Autonomous Shopping Is Redefining the Customer Journey,” May 2026). Same research: AI now drives up to a quarter of referral traffic for some retailers, yet still under 1% of total traffic. Influence is real. Autonomy is not.
Retailers figured this out faster than Google did. When I surveyed top merchants in early 2025, the answer was consistent and blunt: they would use AI and their own data to improve the experience inside their own domain, and they would resist any intermediary that reduced them to price competition. They were not interested in funding a new Google.
The Part Google Is Getting Right
Merchant enablement is genuinely good work, and Google Cloud is where it shows.
Look at Michaels. In May they launched “Ask Mike”, a conversational shopping agent built on Gemini Enterprise for Customer Experience, and by late July it had run close to 75,000 conversations with more than 60% of interactions focused on product discovery. Concept to production grade in six weeks (Google Cloud press release, July 21, 2026). Shoppers are not asking it to buy things unattended. They are asking it to help plan a child’s birthday party, or to find fabric for DIY curtains, and then they buy on Michaels.com, with Michaels as merchant of record, holding the data and the risk.
That is the shape of the thing that works. It is a specialist agent, owned by the retailer, trained on that retailer’s assortment and category expertise, sitting exactly where the customer relationship and the fraud liability already live. Google supplied the model, the infrastructure, and (importantly) the speed. Google did not supply the judgment about what a punch needle pillow project requires.
This is consistent with the argument I made in Merchant Center Becomes the Center: the leverage is in structured merchant data, because that is what an agent actually reads. Feed quality is now a bidding issue. Enablement is the durable business. Sell the picks and shovels.
The Part I Do Not Buy
Where I part company with Google is control. Google wants to be the orchestrator, the standards body, the payment enabler, the authenticator and the referee simultaneously. UCP was co developed with Shopify and marketed as processor agnostic, which is fine, but UBS noted at NRF 2026 that the initial payment method was Google Pay with PayPal to follow (UBS Global Research, “Initial thoughts on agentic commerce announcements & takeaways from NRF 2026,” January 14, 2026). Open standard, house wallet first. We have all watched this movie.
The bigger problem is self attestation (did intent match what was purchased). In the emerging intent frameworks, the agentic platform attests that what landed in the cart matches what the consumer asked for. As I argued in Agentic Data Battle: Intent and again in Agentic Apocalypse, How to Stop It, that is the platform grading its own homework. Intent is not a boolean. “Navy cotton joggers, medium, machine washable, under $60” is five constraints, and an agent can satisfy four, fail one, declare success, and hand the merchant the return. The party that earns the transaction fee should not be the party certifying that the transaction was correct.
I also do not believe the end state is one omniscient Gemini holding every consumer’s data point and every intent with every merchant’s consumer prefs. It will be many specialist agents: retailer agents like Ask Mike, category agents, travel agents, procurement agents, bank agents that hold instrument risk. Orchestration between them is a real and valuable job, and Google may well win it. Adjudication is a different job, and it needs a neutral party, as I discussed in Can Processors Win a Role in Agentic?
Merchants Hold the Cards Right Now
This is the window merchants should be using, because it will not stay open. Merchants own the checkout. Merchants own the fraud loss (all of it in the US, half in the EU). Merchants own the return, the refund, the customer service call, and the lifetime value. Every protocol proposed so far assumes merchant cooperation, which means merchant terms are negotiable today in a way they will not be once one standard achieves default status.
Merchant recommendations – Three things I would do now:
- Build the specialist agent in your own domain, the way Michaels did. Consumers trust it more, and every conversation is proprietary intent data you keep.
- Publish your structured product data aggressively, and treat feed quality as a revenue function rather than a hygiene task. Discoverability by agents is the new retailer imperative.
- Refuse self attested intent. Insist on independent verification of intent to purchase match before you release goods and accept the liability. If you accept a platform’s word today, you will be accepting it in 2030.
Verdict
Google pulling back from Buy for me is not a retreat, it is a correction, and a smart one. Merchant enablement is the better business and the more defensible position. Google Cloud is executing well there. But enablement and control are separate ambitions, and Google is pursuing both at once. Merchants should buy enthusiastically from the first and negotiate hard against the second. In the early innings of agentic commerce, the party holding the checkout and the risk holds the pen. Use it.
“Orchestration between them is a real and valuable job”
Thoughts on Plaid as a potential orchestrator? They’re heading in this direction with the Sierra partnership announced yesterday.
I’m probably not the best person to opine on Plaid given my negative views. To make orchestration work you need to be able to move 3 parties: Merchant, Consumer and Consumer’s Bank. Google is has the largest merchant AND consumer footprint of any company. People pay for services, with service agreements. WHen Google has trouble moving merchants.. its a even bigger lift for any company without the data, history, footpring and reach. The tech in orchestration is easy.. anyone can do it. Moving the parties into an economic model is the challenge. Plaid is not a company deep in agreements. The value in orchestration is unlocked by the company that can operationalize the tech (not provide the API).
Your third recommendation is the one merchants will find hardest and the one that matters most. Independent verification of intent-to-purchase match is the right requirement.
The open question is who can hold it. You point to a neutral party, and in the processors piece to the networks. But the networks earn on the transaction too. If the adjudicator’s economics move with volume, we are back to grading homework, one layer up.
The other half is that the customer has no representative in any of these protocols.
Platform, merchant, network: every party at the table is a counterparty to the customer, not an agent of the customer. Which is why I read the Bain 3-1 trust gap as a measure of a field in which 3rd-party agents carry no credentials and no liability, rather than as a durable customer preference.
Disclosure: I am building in this space, so discount accordingly.
it is a trust function….. and trust has to be where risk is managed. If a bank holds the risk then independent verification will sit with bank (or its agent), if it is with merchant than with merchant or merchant agent. Right now merchants own it all and merchants don’t really care about intent because the are getting the consumer into their site to finish checkout. I a future M2M world (that is not here yet) this needs to be solved.