For years, the purchasing funnel always ended in the same place: a person with their mouse hovering over the «Pay» button. The agency trade This is turning that scenario on its head. Now, the person browsing the catalogue, comparing prices and, in some cases, completing the purchase may well be an artificial intelligence agent acting on the user’s behalf. And this isn’t a laboratory experiment: by 2026, there will already be platforms, protocols and funding behind it. At Vandelay, we’ve been seeing this coming for months, so we’re going to explain it clearly and transparently, with our sources laid out on the table.
What exactly is agency-based trading?
Agency trade (in English, agentic commerce) is a model of buying and selling in which AI agents research, evaluate options, negotiate and complete purchases on behalf of consumers or businesses, sometimes without direct human intervention. This is how IBM defines the term, and it aligns with McKinsey’s interpretation, which describes it as a shift that transforms the purchasing process — hitherto a series of discrete steps: searching, browsing, comparing and buying — into a continuous flow guided by the user’s intent.
The difference compared to a traditional search engine or price comparison site is subtle but huge: the agent doesn’t simply provide you with ten links to choose from, but instead decides (or shortlists) for you based on your preferences. For a brand, this means that the «customer» they need to convince is no longer just a person, but also a system that reads, interprets and recommends products.
Why 2026 is the year it ceased to be just a theory
The most talked-about case is also the most instructive. OpenAI launched its feature Instant Checkout on ChatGPT in September 2025, with the aim of allowing users to make purchases without leaving the conversation. In March 2026, as reported by CNBC, OpenAI shifted its strategy and stopped pushing for direct purchases within ChatGPT to focus instead on product discovery and driving qualified traffic to retailers. The reason was not a lack of interest, but the difficulty of resolving some far from trivial issues: taxation, fraud prevention and real-time inventory synchronisation on a large scale.
Just because the pioneer has taken a step back does not mean the movement has come to a standstill; it means it is getting organised. Other players — Perplexity, Microsoft with Copilot and Google — have launched or announced their own AI-assisted shopping experiences and protocols, relying on payment gateways such as PayPal or Stripe and e-commerce platforms such as Shopify. The industry conversation no longer centres on «whether» retailers will adopt these solutions, but on «how» they can be standardised so that they can do so with confidence.
The figures that explain all this interest
If so many companies are taking action at the same time, it is because the projections are hard to ignore. It is best to treat them for what they are — estimates, not certainties — but the order of magnitude speaks for itself:
- McKinsey estimates that agent-based trading could generate up to 1 trillion dollars in orchestrated retail sales in the United States and amongst 3 and 5 trillion dollars globally by 2030.
- A study by the IBM Institute for Business Value, cited by IBM, states that the 45% for consumers already uses AI at some stage in its purchasing process.
Two figures, two independent sources and the same conclusion: AI-assisted shopping is no longer a niche phenomenon. Although the complete processing of payments by an agent is still in its early stages, the research and recommendation phase is already taking place today, in the hands of users.
What really changes for your brand?
This is the point that really matters to any business. When an AI agent is the one «reading» your catalogue, the quality of your data ceases to be a technical detail and becomes a selling point. An agent isn’t swayed by a pretty picture: it needs comprehensive product details, clear prices, reliable availability and structured trust signals in order to understand and recommend a product. If that information is missing or disorganised, your product simply won’t feature in the conversation.
This ties in directly with two topics we have already covered on the blog. On the one hand, the Generative Engine Optimisation (GEO), which determines whether the AI mentions and recommends your brand. On the other hand, the social commerce and, in general, the development of the agentic marketing, where agents are already playing a part in how campaigns are run. Agent-based commerce is, in a way, the missing piece: it’s not just about AI talking about your brand, but about it being able to buy it.
How to start building your brand without going mad
1. Organise your catalogue and product details
Before you start thinking about agents, check the basics: descriptive titles, complete product details (size, colour, material, compatibility), up-to-date prices and synchronised stock levels. It’s the least glamorous but most profitable part of the job, because it’s exactly what an agent needs to ensure they don’t rule out your product.
2. Structure the information so that machines can understand it
Structured data (schema.org), the feeds well-structured product descriptions and resources such as the llms.txt file They help AI systems correctly interpret what you’re selling, to whom and under what conditions. The more ‘readable’ you are to a machine, the better your chances of appearing in its recommendations.
3. Strengthen your signs of trust
Verifiable reviews, clear returns policies and transparent delivery details don’t just win people over: they are the very same indicators that an agent uses to assess whether they can confidently recommend your brand. Reputation becomes readable through code.
4. Measure and experiment sensibly
Traffic from AI assistants behaves differently to that from a traditional search engine. It’s worth starting to separate it out in your analytics, monitoring how it develops and carrying out controlled tests before making any major decisions. It’s too early to go all in, but it’s too late not to be keeping an eye on it.
Our reading at Vandelay
Agent-based commerce is at a juncture that is both awkward and fascinating: there is enough momentum to take it seriously, yet enough immaturity that no one has yet found the winning formula. OpenAI’s partial withdrawal from direct payment demonstrates this: the sector is learning first-hand what works and what doesn’t. Precisely for this reason, the advantage will not go to those who arrive last, nor to those who take the plunge without a safety net, but to those who have their data, their catalogue and their trust signals ready by the time the infrastructure has fully settled.
Our advice is simple and unspectacular – which is usually the best sort: don’t chase every headline, but start laying the foundations today. If your brand is already clear, trustworthy and well-structured for search engines, it won’t matter which platform ends up dominating. You’ll be on board, no matter who’s doing the driving. And if you’d like us to help you get there, that’s what we’re here for.
Sources
- McKinsey & Company — The agentic commerce opportunity
- IBM — What Is Agentic Commerce?
- CNBC — OpenAI overhauls the shopping experience in ChatGPT after struggling with Instant Checkout
- commercetools — AI Trends Shaping Agentic Commerce in 2026
English (UK) 
Español