Have You Added AI Agents to Your Customer Personas Yet?
I’ve been doing a lot of thinking lately about who, exactly, is using the products I’m asked to assess. It sounds like a strange question, but it’s starting to have an answer that isn’t always a human, and I suspect most boards haven’t put it on the agenda yet.
Most companies still define customer experience as the experience of a human using the product. The personas are human: the busy procurement lead, the cautious IT evaluator, etc. The journey map starts at your homepage and walks through carefully designed navigation towards a beautifully optimised call to action; the whole discipline of customer focus assumes there is a person at the end of the funnel looking at what you built.
That assumption is starting to break, because one of your customer types may now be an agent acting on behalf of your customer. The buyer is still a person, still the one with the budget and the intent; the hands on the keyboard, though, may belong to their AI assistant. If the customer says “find me the cheapest option, check availability and buy it,” their agent may well skip your homepage, your navigation and your CTA altogether. What it wants to know is closer to five things: what can I do here, what data can I access, what actions can I take, what are the rules, and can I transact safely. That’s a different kind of visitor, arriving with the same commercial intent and almost none of the visual engagement your funnel was built to measure.
You can see this happening already in commerce, where Shopify appears to be treating agents as another distribution channel. Last December it made Shopify stores agent-ready by default, discoverable inside ChatGPT, Copilot, etc.,1 and by June a developer could register an agent profile and build the full flow from product search to checkout against a public endpoint.2 Microsoft, for its part, has put a Commerce MCP server into preview that it describes as built around customer intent rather than individual APIs.3 In both cases the agent gets answers directly, and the answers cover the same short list: what it can do, what it can access, and what rules apply.
So the line I’d put to a board is this: your next customer may not be a person; it may be their agent. The follow-on matters more, because if you’re doing customer journey mapping today there should probably be an agent journey beside the human one. That moves the conversation out of MCP and APIs and into customer strategy, which is where I think it belongs; underneath, this is a question about who you serve and how they reach you.
An agent journey looks quite different from a human one; there is very little awareness or consideration in it, and brand preference and emotional resonance barely register. It starts with discovery: can an agent find you at all, through structured data, product feeds or a protocol endpoint rather than your homepage. It continues with access: can the agent read what it needs (prices, availability, etc.) in a form it can use, under permissions that make sense. It ends with transaction: can it complete the action the customer asked for, safely and within the rules the customer set, which is the same consequence and verification question I’d ask about any agent acting on its own. Your existing funnel analytics will probably see very little of this journey, because the funnel measures human attention and this visitor spends almost none; it simply gets on with the job.
The uncomfortable implication is that some of what companies prize most in customer experience becomes close to invisible to this new persona. Beautiful design, persuasive copy, etc. do very little for an agent; what it responds to is structure, meaning clean data, clear rules and honest answers about what it’s allowed to do. It’s a strange inversion, since empathy, design and story tend to win human customers while agents are won by the unglamorous work of well-described products, consistent pricing, dependable APIs, etc. You end up with two personas on very different journeys through the same product.
There’s an ownership angle too, and readers of this blog will know I think ownership is the real issue in most technology questions; it’s the second of the three questions I ask about AI agents. Who in your company owns the agent experience? In most organisations I assess it doesn’t appear to have an owner, because it has no natural home: marketing owns the human journey, sales owns the funnel and product owns the roadmap, so the agent journey ends up sitting in the seams. Unowned capability creates debt for someone later, and here the debt usually arrives as a competitor who made themselves agent-friendly while you were optimising a homepage the new customer rarely sees.
If this sounds abstract, there’s a small test worth running. Ask your team to find your company the way an agent would: skip the brand name in the search bar, ask an assistant to do the job your product does, and see what comes back. It’s usually illuminating, and I suspect it’s a better starting point for the agent journey than most frameworks.
I should be honest about the pace of this, as always. In most sectors the agent customer is a small share of demand today, and it would be premature to rebuild your customer strategy around it. The direction looks clear, though; the platforms have moved first, and in some markets buying agents are already stripping procurement friction out of resale. The companies that treat agents as a persona now, even a small one in the corner of the journey map, are likely to be the ones that aren’t guessing when the share shifts. Boards ask me about AI strategy often, and this is quietly one of the most board-friendly versions of the question, because it skips models and tooling and goes straight to who your customer is.
Your next customer may not be a person; it may be their agent, and a journey map that assumes otherwise is probably missing a lane.