Agentic Commerce 2026: Mid-Year State of the Market

The infrastructure gap is visible in the numbers. Walmart's in-chat purchases converted roughly 3x worse than sessions redirected to its own site. The agent surfaced the demand. The checkout was not built to close it.

That gap is an engineering problem, not a demand problem. The plumbing — checkout, payments, identity, authentication — was designed for a person present at the keyboard. An agent hits each of those gates and stalls.

The result: teams optimizing for human persuasion are polishing a surface fewer buyers ever reach. An agent reads the product feed, not the hero image. It compares attributes, not brand story. When a listing is thin, the agent does not fill the gap. It moves to a cleaner competitor.

Three Structural Shifts Now Underway

1. Structured Data Is Distribution

Structured data is no longer a hygiene requirement. It is a distribution lever. Pages with structured data are cited approximately 3.1x more often in Google's AI Overviews. Research shows 71% of pages cited by ChatGPT and 65% cited by Google AI Mode carry structured data.

An agent evaluating a product listing is less forgiving than a human shopper. Missing size attributes, inconsistent descriptions, or stale pricing do not produce lower conversion. They produce exclusion. The product never enters the shortlist.

This reframes the product data conversation. Data quality used to protect against downstream operations problems. Now it decides whether a machine-driven buyer can see you at all.

2. Trust Is Shifting From a Marketing Question to an Identity Question

The relevant trust question is no longer whether a shopper trusts a brand. It is whether a system can verify the agent at the door.

The gap between the top and bottom of that table — 65% willing to let an agent compare prices, 14% willing to let it spend money — is the trust gap the market is working to close. It will not close through marketing. It closes through KYA infrastructure.

This is the Know Your Agent problem, or KYA. It runs parallel to Know Your Customer, but the subject changes. KYC verifies the person. KYA verifies the agent acting under that person's delegated authority, across three dimensions:

• Identity: Is the agent legitimate, and on whose behalf is it acting?
• Authorization: What is it permitted to do, and up to what transactional limit?
• Behavior: Is it operating within expected parameters, or has it drifted?

Existing fraud detection was built for human sessions, device fingerprints, and static identity checks. Agent activity runs continuously, adapts to detection patterns, and mimics legitimate purchasing behavior.

The case for building KYA now, rather than after autonomous checkout scales, is straightforward.

3. Discovery Readiness and Transaction Readiness Are Two Separate Builds

Being findable by a machine and being able to let a machine transact are distinct engineering problems. Discovery readiness makes a catalog legible to an agent. Transaction readiness allows an agent to act under delegated authority without the system breaking or being defrauded.

The conversion gap is narrowing. Revenue per AI-referred visit grew 84% from January to July 2025 alone. But that same table shows the gap still exists, and that its cause is structural: checkout and payment flows are not yet built for agents. The narrowing is progress. It is not a solution.

Treating discovery readiness and transaction readiness as one task produces false confidence. A business can be fully discoverable and completely unready to accept an autonomous purchase. The line between them needs to be drawn and the sequencing decided before the pressure arrives.

The governing position: any commerce system now serves two buyers simultaneously. The human forms the intent and holds the trust. The agent executes against that intent and enforces the stated constraints. Infrastructure designed for one will fail the other.

Where the Work Sits Now

The decisive engineering is no longer happening at the storefront surface. It is happening at the substrate: the data layer, the identity layer, the instrumentation layer.

The order of operations that has held across agentic commerce deployments — including Hohen Ventures' work building ForgeCart as an autonomous commerce system — is consistent. The constraint is never generating demand. It is verifying the actor and making the catalog readable. Everything structurally expensive sits on that side of the line.

For operators, this translates to four prioritized moves:

1. Audit product data for completeness, consistency, and freshness first. It is the cheapest, highest-leverage intervention available and the one most commonly deferred.
2. Build identity and authorization for a new class of actor. Plan KYA infrastructure alongside KYC before autonomous checkout is common, not in response to a fraud event.
3. Instrument agent traffic separately. Most stacks cannot currently distinguish agent activity from human sessions. You cannot manage a channel you cannot measure.
4. Sequence discovery readiness before transaction readiness. Win discoverability now. Gate autonomous transacting behind verification infrastructure you can trust.

Market Positioning: What Separates the Ready From the Rest

Agentic commerce is not arriving as a single event. It is arriving at two speeds, and the mismatch will not hold indefinitely.

Those projections describe what is available to businesses that treated machine readability as a core requirement early, not as a retrofit.

Discovery is already operating at scale. Autonomous checkout is constrained by trust, identity, and payment infrastructure that is still being built. The current window — where agent-led discovery has outpaced agent-led purchasing — is the cheapest moment to prepare and the easiest to defer.

The businesses that capture agent-driven demand share one defining trait: they treated the machine as a real customer before it was convenient to do so. Clean data, agent-aware identity, and infrastructure that closes the transaction loop without a human present are the common denominators.

The alternative is to keep optimizing for persuasion, and to find that the storefront was polished for a shopper who had already handed the decision to something it was never designed for.

The agent is already reading the catalog. The only open question is whether it can find you in it.