When Your Customer Is an Agent

The most interesting strategy conversations we are having this quarter share a theme: clients asking, some nervously, some eagerly, what happens to their brand when a growing share of purchases is initiated, compared, and completed by software acting on a customer's behalf. The protocols are early, the standards fights are unresolved, and the demos are ahead of the deployments. But the direction is no longer seriously contested, and "wait and see" has quietly become a strategy with a price tag.

Here is our working view, stripped of both the hype and the doom.

What actually changes

When an agent shops, the messy middle of the funnel compresses. The agent does not experience your homepage takeover, does not linger on your product photography, does not feel the checkout's micro-delights. It evaluates structured claims: price, availability, specifications, terms, reviews, and, critically, whatever reputation signal it can gather about whether your brand keeps its promises.

Three consequences follow.

• Verifiable claims beat evocative ones at the transaction layer. "Ships in two days, 2.1 percent defect rate, free returns honored in an average of 3.4 days" is agent-legible. "Uncompromising quality" is noise.
• Operational excellence becomes marketing. An agent that gets burned by a late delivery adjusts, at scale, instantly. Your fulfillment metrics are now brand equity in the most literal sense.
• The default gets powerful. Agents will carry preference weightings, and being a customer's standing default ("always buy my usual from Northwind unless something is 20 percent better") is the new loyalty. Winning the default is cheaper than winning it back.

What stubbornly does not change

Here is where I break from the breathless version of this thesis. Agents are instructed by humans, and humans form preferences the way they always have: through memory, feeling, status, story, and habit. Somebody has to tell the agent what "my usual" means. The desire still has to be manufactured somewhere, and it will not be manufactured in a price-comparison matrix.

If anything, agentic commerce splits brand work into two cleaner jobs. Upstream: build enough meaning and preference that humans specify you by name in their instructions. Downstream: be so operationally impeccable and machine-legible that the agent never finds a reason to override that instruction. Brands strong in only one of these will leak share to brands strong in both.

The analogy we keep using with clients: this is the private-label dynamic replayed at protocol speed. Retailers' house brands won wherever consumer preference was weak and lost wherever it was strong. Agents are about to run that experiment across every category simultaneously.

What to do in 2026, concretely

You do not need to bet on any particular protocol winning to act now. The no-regret moves are clear.

• Audit machine legibility. Structured data, consistent entity naming, published terms, accessible specs. If an agent cannot verify it, it does not exist.
• Instrument your operational truth. Know your real delivery times, return friction, and defect rates, because agents will know them whether you do or not.
• Invest in named preference. Measure how often customers ask for you by name, in search, in assistants, in the wild. That number is your future default share.
• Decide your agent posture. Will you expose transactional endpoints early, or force agents to scrape? Early cooperators are earning trust weightings that will be expensive to buy later.

The honest summary: agent-to-agent commerce is still partly speculation, and anyone selling certainty about its timeline is selling something else too. But the capabilities it rewards, verifiable truth, operational excellence, and human preference strong enough to survive delegation, are worth building regardless of when the protocols settle. That is the definition of a good strategic bet. Make it calmly.