What is Agentic Commerce?
AI agents are becoming the new shoppers. Here's what that means for your business.
The Paradigm Shift
For decades, e-commerce has been designed around human shoppers: people who browse visually, click buttons, fill forms, and make decisions based on what they see on screen. But a fundamental shift is underway.
Today, millions of people are asking AI assistants like Claude, ChatGPT, and Gemini to help them shop. "Find me running shoes under €150" or "What's the best laptop for video editing?"
These AI agents need to discover products, understand specifications, compare prices, and potentially complete purchases, all programmatically. Websites built only for human eyes are becoming invisible to this new class of shoppers.
Traditional vs Agentic Shopping
- 1Browse pages visually
- 2Read product descriptions
- 3Click "Add to Cart" button
- 4Fill checkout forms manually
- 5Enter payment details
- ✓Query product APIs directly
- ✓Parse structured data (Schema.org)
- ✓Add items via cart API
- ✓Submit orders programmatically
- ✓Use secure payment tokens
A Real-World Scenario
"Hey Claude, I need new running shoes. I like Nike, size 10, and I want to spend less than €150. Can you find something and order it for me?"
To fulfil this request, an AI agent needs to:
If your website only works for humans clicking buttons, the AI agent cannot help this customer buy from you. They'll go somewhere that's AI-ready, likely your competitor.
Under the Hood: How Agents Browse, Buy and Pay
The scenario above sounds simple enough. But what's actually happening behind the scenes? Here's what a well-equipped AI agent does, step by step, and why each step matters for your business.
The agent starts by looking for a commerce manifest: a single file that tells it everything about your shop's APIs, accepted payment methods and policies. Think of it like a menu for machines. If that file isn't there, the agent has to guess, and it'll often give up and try a competitor instead.
Once it finds your catalog API, it reads the structured product data. Not the nice photos and marketing copy (agents can't appreciate either of those), but the machine-readable details: SKUs, dimensions, stock levels, prices per variant. This is why Schema.org markup matters so much.
Notice that adding to cart and paying are two separate API calls. This is deliberate. The agent creates an order first to lock in the exact total (including tax and shipping), so the user has agreed to a specific number before any money moves. It's the same logic as a human reviewing their basket before tapping "Buy", but done programmatically in milliseconds.
Sites without a cart or orders API force the agent to navigate checkout forms like a very clunky robot. This almost always fails. CAPTCHA challenges, session timeouts and JavaScript-rendered forms all block agents stone dead.
This is where it gets interesting. Agents can't type card numbers into forms, and they definitely shouldn't be storing raw card details. Instead, there are three emerging approaches to agent-native payment.
- 1User's card is vaulted securely by VGS or their bank. The actual card number never leaves that vault.
- 2The agent receives a network token: a single-use alias for that card, scoped to one merchant.
- 3A short-lived cryptogram is generated for this exact transaction.
- 4Merchant charges it like a normal card. Works on any existing payment terminal.
- 1Agent sends a payment request to the merchant's pay endpoint.
- 2Merchant responds with HTTP 402, stating the amount, currency (e.g. USDC) and wallet address.
- 3Agent settles the stablecoin payment on-chain in seconds.
- 4Agent retries the request with proof of payment. Order confirmed.
- 1The user pre-authorises a spending wallet with Skyfire, setting limits (e.g. max £200 per day).
- 2The agent attaches a single signed token to its request header.
- 3The merchant verifies the token in milliseconds. No redirect, no 3DS popup.
- 4Payment is settled and the agent is identified at the same time.
Why This Matters Now
The businesses that adapt now will capture this emerging market. Those that wait will find themselves invisible to a growing segment of consumers who shop through AI.
What Makes a Site AI-Ready?
AI-ready websites share three fundamental qualities. At Aidō, we developed the D/U/T Framework to measure these:
THE PAYMENTS LAYER
A new generation of payment infrastructure is emerging specifically for agentic commerce. nekuda provides merchants with a hosted wallet SDK so AI agents can complete checkout using securely stored credentials. PayOS works at the network level, using Mastercard Agentic Tokens and Visa Trusted Agent Protocol, so any merchant on standard card rails is instantly compatible, with no integration required.
Read more about agentic payment protocols →Common Blockers
Many e-commerce sites unknowingly block AI agents. Here are the most common issues:
-
Blocking AI crawlers: robots.txt rules that prevent GPTBot, ClaudeBot, and other AI crawlers from accessing your site.
-
No structured data: product pages without Schema.org markup; AI agents see garbled text instead of clear product info.
-
JavaScript-only content: products that only appear after JavaScript runs; most AI crawlers don't execute JavaScript.
-
No APIs: no programmatic way to search products, manage carts, or complete checkouts.
-
CAPTCHA everywhere — Bot protection that blocks legitimate AI agents along with malicious bots.
Where Does Your Site Stand?
Our free scan analyses your website across all three dimensions and shows you exactly what to improve.
Join the Beta