Agentic commerce: what happens when the buyer is a bot

The next buyer walking into your store may be software. Not a scraper stealing prices — an AI agent with a human's intent, a human's budget, and a task: "find me the best option and buy it." Agentic commerce is early, but it's no longer hypothetical, and stores are split between ignoring it and over-preparing for a version that doesn't exist yet. Here's the grounded middle.

What agentic buyers do today, concretely

Three distinct behaviors are already observable on real stores:

Research fetches. An assistant (ChatGPT-User and friends) pulls your product page mid-conversation because a human asked "is this jacket waterproof?" or "which of these two should I buy?" The agent reads your page and answers — your product copy is being interviewed without you present.

Comparison sweeps. Agentic browsers — AI driving a real browser — visit several stores in one task, extract prices, shipping, and return policies, and report back a shortlist. They execute JavaScript and look almost human in analytics, which is exactly the problem: your conversion rate now includes visitors who were never going to "convert" in the human sense but decided whether their human ever sees you.

Assisted checkout. The frontier: the agent completing the purchase — filling carts and forms with the human approving the payment. Early, fragile, but every major AI lab shipped or previewed a version of it, and checkout flows that break agents (CAPTCHA walls, JS-only pricing, forced account creation) are already measurably hostile to it.

Why this scrambles your metrics first

Before agentic commerce changes your revenue, it changes your data. An agent that browses five product pages in twelve seconds and leaves isn't a bounced human — it may have shortlisted you. A "visitor" that never scrolls isn't disengaged — it read the DOM, not the viewport. Stores that can't tell humans from agents are already optimizing against corrupted numbers: ad-audience pixels firing on bots, A/B tests polluted, retargeting budgets spent on software.

That's the practical first step, and it costs nothing strategic: classify the traffic. Every downstream decision — from "is this channel real for us" to "should agents see different content" — needs that split to exist.

Preparing without guessing

Make your store machine-legible. Prices, availability, shipping, returns in plain crawlable HTML plus accurate Product structured data. Agents recommend what they can parse; ambiguity loses shortlists. This is the same work AI search rewards — one investment, two channels.

Don't reflexively block. A blocked agent doesn't tell its human "this store is protective of its data." It says "I couldn't access that store — here are three alternatives." The calculus differs from content scraping: for commerce, the agent is the customer's hands.

Watch the checkout funnel for agents specifically. Where do they stall? Every wall that stops an agent with purchase authority is now a form of cart abandonment. You don't need to redesign checkout today — you need the measurement in place so you know when it starts mattering, per week, in your store, not in a think piece.

Shopify stores: this measurement is a theme app-embed toggle, not a build.

The honest timeline

Agentic purchases are a small fraction of commerce today, and anyone selling you urgency is selling something. But the research and comparison behaviors are already in your traffic — silently, inside your "Direct" and your bounce rate. The stores that win the transition won't be the ones that predicted the timeline; they'll be the ones that could see the shift in their own numbers the week it happened.