Risk factors are the best agentic commerce adoption index

Tapestry's new 10-K names AI shopping assistants as a competitive risk for the first time, while halving its total AI mentions. Filings are a better adoption signal than surveys.

By XAgent Team · 2026-08-14

Yesterday Tapestry, Inc. filed its annual report for the fiscal year ended June 27, 2026, and it contains a paragraph that was not in last year's: "customers are increasingly using AI shopping assistant tools to discover products, compare options, and make purchase decisions. Use of these AI tools could transform commerce, including in ways that we fail to anticipate, and affect our ability to efficiently attract potential customers to our digital platforms and retain our customer base." If you want to know how far agentic commerce has actually travelled, this kind of sentence is a better instrument than any vendor survey, and almost nobody is reading it.

Why a risk factor is a strong signal

Marketing copy is cheap and earnings-call optimism is free. Risk-factor language is neither. It is drafted by securities lawyers, reviewed by auditors, and signed by officers with personal exposure if the disclosure turns out to have been inadequate. The professional instinct in that room is to delete, not to add — every additional sentence is a new thing you have told the market you are worried about.

So a first-time appearance is informative in a way an interview is not. Someone inside the company assembled enough evidence that a channel shift was underway to justify telling shareholders about it, and the lawyers agreed the risk was real enough to warrant disclosure. That is a higher evidentiary bar than "an executive said it on a call" — which, as we argued about this earnings season's agentic numbers, is where nearly all of the industry's data currently lives.

The detail that rules out AI-washing

The obvious objection is that every company is bolting AI language onto every document right now, so a new AI paragraph means nothing.

Tapestry's filing kills that objection on its own numbers. We counted the terms in both years' documents directly from the filings:

Total mentions of artificial intelligence went down by half. This specific channel-risk language went in anyway. That is the opposite of buzzword inflation — the document got less AI-heavy overall and more specific about one thing.

The second insertion is in the list of competitive factors, where Tapestry now names "competing with AI-enabled shopping tools to help find products, compare prices, and make purchase decisions" alongside conventional items like sourcing and trademark protection. AI shopping tools are being filed as a competitor, not as a technology trend.

What Tapestry does not say is as interesting

The company's single use of the word "agentic" is not about agents buying from it. It appears in the business section, describing its own internal deployment: "we continued to advance our artificial intelligence including predictive, generative and agentic models, as well as machine learning across key areas such as data analytics, planning, marketing, customer acquisition, personalization, pricing and product creation."

So agentic is a technology Tapestry runs. AI shopping assistants are a thing that happens to Tapestry. Nowhere in the filing is a third-party assistant treated as a channel the company could sell through — only as an intermediary that could stand between it and its customers.

That asymmetry is common and it is a strategic tell. A retailer that models agent traffic purely as an acquisition risk will spend on defending direct traffic. A retailer that models it as a channel will spend on being executable by the agents that arrive. Those are different budgets, and right now the disclosure language suggests most brands are in the first camp.

How to run this index yourself

The method is cheap and repeatable, and it produces evidence rather than opinion.

Pull a company's current annual report and the prior year's from EDGAR. Search both for the same set of phrases: AI shopping, shopping assistant, AI-enabled shopping, agentic, conversational commerce, AI agent. Record first appearances, and — this is the part that gives the signal its strength — record whether total AI mentions rose or fell in the same period. A first appearance against a falling AI baseline is a real finding. A first appearance inside a doubling of AI language is noise.

Then read where in the document it sits. Competitive factors and risk factors mean the company sees a threat. Business and strategy sections mean it sees a capability. The two rarely appear together yet.

Do that across a sector and you have an adoption curve built from documents executives signed, months ahead of anyone quoting a percentage out loud.

What's next

Watch for the first filing that puts AI shopping assistants in the opportunity column — a strategy section describing agent channels as a route to customers rather than a risk to acquisition. That inversion will be a more meaningful milestone than any protocol launch, because it will mean a public company has decided the channel is worth spending against.

Until then, the useful reading of Tapestry's filing is narrow and worth taking seriously: a major branded retailer has told its shareholders, in writing, that AI shopping tools could affect its ability to acquire and retain customers. The companies that will do well out of that shift are the ones that make themselves easy to buy from rather than hard to route around.

If you would rather appear in that column as a channel than as a risk, list your store on XAgent and let the open execution market make your catalog quotable, authorizable and executable by the agents already shopping it.