A shopper can now ask an AI assistant to find a quiet laptop under a certain budget, compare the options and complete the purchase, without ever opening a retailer’s website. That capability moved from demo to daily behaviour faster than almost anyone in retail expected. Adobe measured a 693% increase in AI-driven traffic to retail sites over the 2025 holiday season, and the momentum carried into 2026: AI traffic grew another 393% year on year in the first quarter, and by March it converted 42% better than non-AI channels.
Most of the conversation about agentic commerce happens in marketing terms: visibility, recommendations, the new funnel. As COO, I see it differently. When software starts buying, the deciding factor is operational. Agents purchase from the retailers whose data, availability and fulfilment they can read and trust. Everyone else is invisible.
What is agentic commerce?
Agentic commerce is online buying in which an AI agent handles part or all of the transaction on a person’s behalf: discovering products, comparing options, filling a basket, completing payment, then tracking, returning or reordering. The person sets the intent and the limits; the agent does the work.
It’s already commercially significant. Salesforce estimated that AI agents and generative AI tools influenced more than 20% of global online retail sales during the 2025 holiday season. Whatever the precise share, the direction is clear, and the shoppers arriving through these channels are unusually valuable: Adobe’s data shows they spend 48% more time on site and view 13% more pages than visitors from other sources.
Agentic commerce vs traditional e-commerce
The comparison matters because it shows how much of the existing e-commerce playbook stops working when the buyer is software.
| Traditional e-commerce | Agentic commerce | |
| Who navigates | A human browsing your storefront | An agent querying many retailers at once |
| Where discovery happens | Search results, ads, your category pages | Inside the buyer’s assistant |
| What persuades | Design, copy, imagery, urgency | Structured, accurate, machine-readable product data |
| Checkout | Your flow, your conversion optimisation | A protocol-mediated transaction, often completed in chat |
| Loyalty | Won on brand experience | Won on being reliably correct: price, stock, delivery promise |
The heart of the shift sits in the third row. Traditional e-commerce optimises an interface to persuade a human. Agentic commerce exposes structured truth for a machine to evaluate. An agent doesn’t respond to a hero banner. It responds to complete attributes, live availability and a delivery promise the retailer actually keeps.
The agentic commerce protocol, and the standards race
For agents and merchants to transact safely, they need a shared language, and that’s what the Agentic Commerce Protocol (ACP) provides. Co-developed by Stripe and OpenAI and launched in September 2025 alongside Instant Checkout in ChatGPT, ACP is an open standard that lets a merchant integrate once and sell through any compatible agent. Crucially for retailers, the merchant remains the merchant of record, keeping control over what’s sold, how the brand appears and how orders are fulfilled. On the buyer side, OpenAI’s design keeps users in control, with explicit confirmation steps and encrypted payment tokens authorised only for specific amounts and merchants.
ACP is no longer alone. Google launched the Universal Commerce Protocol in January 2026 with retail partners including Shopify, Target and Walmart, and by April its governing council had drawn in Amazon, Meta, Microsoft, Salesforce and Stripe itself. Card networks are building their own agent payment rails in parallel.
My advice to operations and commerce leaders is to treat the standards race the way we treat any vendor landscape that hasn’t settled: avoid betting the stack on a single protocol, and invest in the things every protocol demands. Clean product data, real-time availability, and integration architecture flexible enough to expose both. Those investments hold their value whichever standard wins.
The operational foundation agents depend on
Here’s the part I care most about, because it’s where agentic commerce succeeds or quietly fails: an agent is only as good as the operational data it acts on. If your product information is incomplete or your availability is wrong, an agent doesn’t persevere the way a determined human might. It simply buys elsewhere.
Our work with Maxeda DIY Group, the largest DIY retailer in the Benelux, shows what building that foundation actually involves. Maxeda serves more than 1.5 million customers weekly across 330+ stores, each carrying over 60,000 products, and the most-asked question in any of them is “where can I find product X?”. Answering it at scale meant solving a data problem first. We built a Product Locator platform on a store-first principle: local teams describe locations in their own human terms, staff assign products to aisles by scanning them with a handheld app, and a full store can be set up in three to four hours. The result is a dynamic base of over 17 million product-location combinations that feeds the website, the mobile app and a conversational AI that answers location questions instantly.
The chatbot is the visible part. The reason it works is the unglamorous operational layer underneath: data owned by the people closest to it, structured so machines can consume it, kept current as shelves change. That is precisely the shape of work agentic commerce will demand from every retailer, because an agent asking your systems “is this in stock, and when can it arrive?” needs the same trustworthy, machine-consumable answers.
What retailers should do now
- Audit your product data as if a machine were the customer. Complete attributes, consistent taxonomy, accurate stock. This is the single highest-leverage investment.
- Fix machine readability. Adobe’s analysis found retail product pages average only 66% visibility to AI models, meaning a third of the content agents need is effectively invisible.
- Set confirmation thresholds deliberately. Low-value reorders can run autonomously; high-value or irreversible purchases should require explicit human sign-off.
- Track the protocols without committing your stack to one. Build the structured data layer they all share.
Key takeaways
Agentic commerce means an AI agent, carrying a person’s intent and payment authority, completes purchases on their behalf. It differs from traditional e-commerce at the root: machine evaluation replaces human persuasion, so structured, accurate operational data becomes the storefront. Protocols like ACP and UCP are settling the plumbing, and the retailers who win early will be the ones whose operations were ready before the channel matured.
If you want an honest assessment of whether your commerce data and systems are ready for buyers made of software, our e-commerce team does exactly this work, from data foundations to AI-powered shopping experiences. The agents are coming either way. Ready is better.
