"Headless" used to describe an architecture. If you decouple the backend from the presentation layer, you can render the same commerce engine in different interfaces. Engineering teams chose to work that way.
Soon it will describe a market condition too. When customers delegate to agents, the presentation layer you spent years polishing gets parsed instead of seen. That choice is made by your buyers rather than your architects, whether or not you designed for it.
The service starts to matter more
Platform shifts tend to move the point of competition. The web made storefront merchandising less important than logistics and selection. Mobile made desktop feature depth less important than being available at the moment of intent. In both cases, incumbents could keep winning on the old surface while the new one started to matter more.
Agents move the point of competition again, this time away from the interface and toward the service itself. When an agent chooses on a customer's behalf, the most beautiful experience matters less than one it can understand and complete. The facts need to be findable and trustworthy. Policies need to be written where a machine can read them. A non-human visitor needs to be able to finish the flow, with availability and pricing stated plainly.
In practice, an agent tends to prefer the service it can complete reliably over the brand it would otherwise recognize.
An agent's preference for you is not brand affinity. It is a completion rate.
Agent experience becomes a discipline
Here's a prediction we feel reasonably good about: within a few years, agent-facing quality becomes a measured discipline, much as human-facing performance did. Core Web Vitals turned page speed from a vague virtue into a number with an owner. We expect the agent side to get an equivalent. Businesses will track task completion, machine legibility, time-to-answer, and error honesty, because when something starts to affect revenue, somebody usually gets asked to measure it.
Organizations working this way will look a little strange by today's standards. They'll review their sites as an agent reads them, not only as a customer sees them. A broken agent flow becomes a sev-one instead of a curiosity. Non-human visitors show up in A/B tests. "Does it demo well?" stops being the only question that matters.
The big caveat
None of this is actionable yet, and that's the main caveat. You can't improve a completion rate until you can compute one. Today, those numbers mostly don't exist because the stack labels every agent "bot" and the evidence never surfaces. A lot of the industry conversation has already moved on to redesigning the experience. That feels premature when almost nobody can say what agents actually do on their site today.
Our timeline could easily be wrong too. The measurement discipline might show up in two years or in seven. A lot depends on how many agents are doing real commerce rather than just fetching answers. We still doubt the head stays on; we just don't know when that change becomes obvious.