Shopify’s president Harley Finkelstein told TechCrunch this week that AI shopping agents will “change everything” about e-commerce. The claim is large. The structural threat it describes — to fashion brands in particular — is specific and underexamined.

An AI shopping agent is software that browses, compares, and purchases on behalf of a consumer. The shopper states a goal (“black ankle boots, leather, under $200, available by Friday”) and the agent handles the rest: searching across merchants, evaluating options against stated criteria, completing checkout. Shopify is positioning its platform to be the infrastructure these agents transact through.

The channel shift no one in fashion is modelling

Fashion retail has spent two decades building direct-to-consumer channels designed to control the brand experience. Lookbooks, editorial landing pages, curated navigation — all assume a human is browsing. AI agents don’t browse. They query, filter, and rank.

If agents become a meaningful purchase channel, the consumer never sees the product page. They see a shortlist, selected by an algorithm optimizing for the buyer’s stated constraints. Brand storytelling, which drives a disproportionate share of fashion margin, drops out of the transaction entirely.

Shopify’s bet makes strategic sense for its merchants: if agents drive traffic, Shopify wants to be the checkout layer they land on. For fashion brands built on Shopify — and there are over two million merchants on the platform — the question is whether their product data is structured enough for agents to surface them at all.

What the counterargument misses

The standard objection: consumers won’t delegate fashion purchases to software because fashion is emotional, aspirational, identity-driven. This is correct for a subset of fashion buying. It is wrong for the largest categories by volume.

Basics, workwear, children’s clothing, athletic gear — these are price-sensitive, specification-driven, and low on brand attachment. They are the purchases an agent handles well. Bain & Company forecasts that agentic commerce could account for 15–25% of all U.S. e-commerce by 2030, between $300 billion and $500 billion, with specification-driven categories adopting first and apparel following as consumer trust grows. Even at the conservative end of that range, the volume of fashion transactions routed through agents represents a channel with no brand storytelling, no visual merchandising, and no loyalty program influence.

Amazon is already drawing the legal boundary

The tension is visible in Amazon’s legal offensive against Perplexity. A federal court ruled on March 10 that Perplexity’s Comet browser — an AI agent that logs into Amazon accounts on behalf of consumers — violated the Computer Fraud and Abuse Act by accessing the platform without Amazon’s authorization. Judge Maxine Chesney drew a distinction that matters beyond this case: a consumer giving Comet their login credentials does not constitute authorization from Amazon itself. Amazon, which accounts for roughly 40.5% of U.S. e-commerce (eMarketer, 2025), is treating agent access as unauthorized computer intrusion under federal law.

Shopify is building in the opposite direction. It co-developed the Universal Commerce Protocol with Google, Walmart, Target, and others — an open standard that gives agents a structured way to discover, negotiate, and complete purchases across participating merchants. Fashion brands selling on both types of platform now face a split discovery problem: on open infrastructure, agent visibility depends on structured data and protocol adoption; on closed platforms, the question is jurisdictional before it is technical.

The data readiness problem

Agent-driven commerce rewards structured, machine-readable product data: precise sizing, material composition, care instructions, inventory status, delivery estimates. Most fashion product catalogs are built for humans. Descriptions are evocative (“a flowing silhouette in midnight blue”) rather than precise (“100% silk, 118cm length, relaxed fit, navy #191970”).

A recent arXiv paper, “AgentDrift” (March 2026), studied tool-augmented LLM agents acting as financial advisors when their data sources become corrupted. The finding: standard ranking metrics showed recommendation quality was “largely preserved” — while risk-inappropriate products appeared in 65–93% of turns. Agents never questioned the corrupted data across 1,563 contaminated turns.

The paper studied finance, not fashion. But the structural problem transfers directly. An AI shopping agent pulling product data from a fashion catalog faces the same failure mode: if the data is ambiguous or wrong, the agent will confidently recommend the wrong product, and no conventional quality metric will catch it. There is no product page to correct the impression. The agent’s interpretation is the only touchpoint.

What fashion retailers should do now

Three actions separate the prepared from the exposed.

First, audit product data for machine readability. Every SKU needs structured attributes — not just for search engines, but for agents that parse specifications against buyer constraints. Colour, material, fit, size in standardized formats.

Second, monitor agent traffic. Shopify merchants can already see bot traffic patterns in their analytics. The question is whether anyone is segmenting agent-initiated sessions from human ones and measuring conversion separately.

Third, watch the legal line. The Amazon v. Perplexity injunction will set precedent for agentic commerce. If courts uphold the principle that platform authorization is required for agent access, brands on closed platforms gain protection but lose the channel. If appeals narrow the ruling, brands without agent-optimized data lose visibility in a growing segment.

The momentum is already measurable

Shopify merchants saw 14x more agent-driven orders between January 2025 and January 2026. Google and Shopify have shipped the protocol layer. Bain projects the agentic commerce market at $300–500 billion by 2030. The infrastructure is being poured while most fashion brands are still debating whether to update their product feeds.

Fashion’s current tech stack was built for human eyes. The next class of customers will have none. Whether a brand captures this channel depends on a question that has nothing to do with storytelling: is your product data readable by a machine that has never seen your lookbook?