Agent commerce is what happens when AI agents move from answering shopping questions to doing commerce work: finding options, comparing tradeoffs, preparing carts, drafting merchant actions, and sometimes handing off an approved purchase.
That sounds like ecommerce with a smarter chatbot. It is not. The shift is that the interface is no longer only a website, a search result, or an ad. The buyer may ask an agent for an outcome. The agent then decides which catalogs, merchants, policies, and checkout paths deserve attention.
For a service business, the useful question is not “will AI agents replace websites?” The useful question is: when a buyer’s agent asks the market who can solve this problem, will your business be legible, trusted, and ready to answer?
What is agent commerce?
Agent commerce is commerce where an AI agent performs part of the buying or selling workflow on behalf of a person or business. The agent may discover products, compare offers, answer fit questions, assemble a cart, prepare a quote, or complete a transaction after approval.
AWS describes agentic commerce as online shopping automated by AI agents, from natural-language intent through product discovery, comparisons, price monitoring, and potentially purchasing. The practical version is simpler: the customer describes the job, and software does more of the path between intent and order.
There are levels of autonomy. At the safe end, an agent only recommends options. In the middle, it fills a cart or drafts a quote for review. At the risky end, it buys, changes listings, issues refunds, adjusts pricing, or spends money without a human checkpoint. Most real businesses should begin in the middle: let the agent do the searching and structuring, then require a clear approval surface before money moves.
That is the first honest catch. Agent commerce is promising because it removes friction. It is dangerous for the same reason.
How does agent commerce change discovery?
Agent commerce changes discovery by making machine-readable clarity more important than persuasion alone. A buyer’s agent does not browse the way a human does. It looks for structured facts it can compare: product details, service areas, pricing rules, availability, return policies, support promises, reviews, and proof.
Shopify’s agentic commerce guide gives the merchant version of the same advice: complete product data, standardized fields, clear titles, schema markup, shipping pages, return pages, FAQs, and reviews all make a store easier for agents to understand.
For a service SMB, translate that beyond retail. Your site should make these things easy to extract:
- what you do and do not do
- where you serve
- who you serve best
- what happens after a lead submits a form
- what information you need to quote or schedule
- what proof exists that you can deliver
- how a human can take over
This is still marketing, but the medium changes. You are not only convincing a person. You are supplying enough clean context that an agent can safely recommend you to that person.
What happens at checkout?
Checkout is where agent commerce stops being a content trend and becomes an operations problem. The moment an agent can pay, book, order, refund, discount, or update an account, the business needs rules.
Anthropic’s commerce-agents reference blueprint is useful because it does not pretend the agent should own the whole transaction. Its shopping-agent demo searches, compares, fills carts, and answers policy questions, but the repository states that checkout renders the cart for the host to complete. On the merchant side, writes are staged until a person approves them.
That pattern is the sober one: agent does the work, host system keeps authority, human approves the irreversible step.
OpenAI is moving in the same general direction from the advertising side. Its Sponsored Agents announcement describes clearly labeled business-sponsored conversations after a user clicks an ad, distinct from ChatGPT’s independent answers. It also ties the ad platform into Shopify and HubSpot so merchants can work from existing ecommerce and CRM systems.
The direction is clear. The sales conversation, product explanation, campaign work, and checkout path are moving closer together. The catch is also clear: if the system cannot show what was said, what was recommended, and who approved the next step, the business has created a trust problem, not leverage.
Why does agent identity matter?
Agent identity matters because merchants need to know who is acting before they grant access, accept payment, or trust instructions. A human account is not the same as an unidentified agent operating through that account.
Cloudflare names the issue directly in its Wallets announcement. Agents struggle to onboard to APIs and services because they lack a stable identifier and a native way to pay. Cloudflare’s proposal gives agents optional human-readable wallet identities tied to a Cloudflare account, so merchants can decide how to treat known and unknown agents.
Payment networks are working on the same problem. Ant International, Mastercard, and Visa announced a Know-Your-Agent interoperability effort focused on operator traceability, shared certification requirements, and continuous transaction monitoring. The point is not technical decoration. It is accountability.
A business needs to know:
- which agent made the request
- which person or organization operates it
- what authority the user gave it
- what limits apply to the transaction
- where the receipt and audit trail live
Without those answers, agent commerce becomes a dispute machine.
What should service businesses prepare now?
Service businesses should prepare for agent commerce by making their offer legible, their intake structured, and their approval rules explicit. You do not need a full shopping agent on day one. You need the rails that make an agent useful instead of loose.
Start with the public layer. Your website should state the offer clearly, answer the practical objections, and expose enough structured information for an AI system to understand fit. If referrals slowed and search took over, this is the next turn of the same screw: buyers will increasingly ask AI systems to filter the market before they call anyone.
Then fix the intake layer. A lead form that only says “contact us” is weak for agent commerce. The better pattern is a guided intake that captures constraints: location, timeline, budget range, service need, urgency, photos or files when relevant, and permission to follow up. That gives both humans and agents better raw material.
Finally, define the authority ladder:
- The agent may read public pages and FAQs.
- The agent may answer fit questions from approved knowledge.
- The agent may draft a quote, estimate, campaign, or follow-up.
- The agent may place the draft in a human inbox.
- The agent may send, book, buy, or change records only after approval.
That ladder is boring. Boring is good here. Boring is what keeps the system useful when real customers, real money, and real obligations enter the workflow.
Where does agent commerce fit with marketing?
Agent commerce makes marketing more operational. The winning business is not just the one with the sharpest headline. It is the one whose claims, data, policies, reviews, and response process are easiest for both humans and agents to verify.
This is why agent commerce belongs beside AI search optimization, lead routing, CRM hygiene, and agentic systems work. A buyer’s agent cannot recommend what it cannot understand. A merchant agent cannot safely act where the business has not defined rules. A sales conversation cannot scale if no one can see the receipt trail.
For BBH’s world, the practical path is not “replace your funnel with an agent.” The path is cleaner:
- make the business visible to AI-assisted discovery
- turn messy intake into structured context
- let agents prepare the next best action
- keep approval close to money, reputation, and customer promises
- measure whether the system reduces work or quietly creates more review load
That last point matters. If agent commerce only moves the burden from clicking pages to checking agent mistakes, it has not helped. The standard is not novelty. The standard is fewer dropped leads, clearer buyer fit, faster response, and better receipts.
The bottom line
Agent commerce is the buying journey becoming agent-readable and agent-actionable. AI agents will help people find, compare, question, and eventually buy with less manual browsing.
The businesses that benefit will not be the ones that shout “AI” the loudest. They will be the ones with clean offers, structured data, visible policies, fast handoffs, and strong approval rails.
Prepare for the agent, but do not worship the agent. The goal is still a better human decision on the other side.