Your buyer's agent is not any of the four publics

Gartner projects that by 2028, 90% of B2B purchases will be intermediated by AI agents, channeling more than $15 trillion through automated exchanges. That number gets cited in decks as a growth story. It is a design problem.

Every B2B product serves four publics. The user who operates it daily. The buyer who signs the contract. The regulator who constrains what it can do. The ecosystem of integrators, resellers, and adjacent tools that determine whether the product survives in context. Most product teams design for the user, occasionally remember the buyer, and discover the other two when something breaks.

An AI procurement agent collapses all four into a single API call. It does not experience any of them.

The agent does not feel the UI that the user loves. It skips the demo that convinces the buyer. It never reads the compliance page that satisfies the regulator. It sends a structured query, receives structured data, scores, and recommends. The product team that spent years designing four separate human-facing surfaces is now invisible to the entity actually making the decision.

This is already measurable. The 2X AI Innovation Lab's AI Visibility Index found that 96% of the 70 B2B companies they analyzed were functionally invisible in early-stage AI-driven discovery. The reason is structural. Their marketing infrastructure was built for a human who types a query, clicks a link, and reads a page. The agent does none of those things.

The supply side is behind. Deloitte Digital reports agentic AI adoption among B2B suppliers at just 24%, compared to 38% on the buyer side. Most suppliers have their IT budgets absorbed by ERP upgrades. They are modernizing the plumbing while the storefront becomes irrelevant.

The surface collapse

When a human buyer evaluates your product, each public gets its own surface. The user gets the product experience. The buyer gets the pricing page, the case study, the sales call. The regulator gets the compliance documentation. The ecosystem gets the API docs and the partner portal. These surfaces evolved separately because each public has different needs and different decision criteria.

The AI agent flattens this into one question: does this product meet the specification, at the right price, within the constraints, and can I integrate it? That question requires structured, machine-readable answers. freeCodeCamp's analysis states it plainly: products accessible only through a UI are excluded from agent-driven procurement, and brand equity has no equivalent in AI evaluation.

BigCommerce makes the commercial version concrete. When an AI buyer agent authenticates and queries your catalog, it must receive contract-aware pricing for that specific account. If your platform cannot serve that to an API request, you do not exist. The buyer public and the user public, which product teams have spent decades separating into distinct experiences, must now resolve into a single programmatic response.

The agent is not a fifth public

The tempting move is to add a surface. Build an agent-facing experience alongside the four human-facing ones. Product Leaders Day India framed this as "B2A" design: logic, structured data, and permission protocols instead of emotion and visual hierarchy.

That framing understates what happened. The agent replaced the first three publics. The user, buyer, and regulator all delegated their judgment to it. Only the ecosystem public remains independently human, because integration partnerships still require relationship negotiation that agents cannot yet perform.

This changes the product team's job. Instead of designing four experiences and hoping each public converts, you design one machine-readable surface that encodes the needs of three publics at once, while maintaining a human relationship layer for the fourth. Commercetools reports that 20% of B2B sellers will be forced to engage in agent-led quote negotiations this year alone.

The teams still building buyer journeys, onboarding flows, and regulatory landing pages are doing real work. That work serves the fraction of purchases that remain human-led. For the rest, the product is its structured API surface and its contract-aware pricing logic. The demo does not matter if the agent never requests one.

The four publics framework was designed to help teams see audiences they were ignoring. The new problem is harder. Three of the four publics still exist, still have needs, still regulate and purchase and use. They just stopped showing up. The agent arrives with all three sets of requirements in a single call and experiences none of them. If your product cannot answer that call, the three publics you spent years designing for will choose a competitor they never evaluated themselves.

Sol's analysis is precise for private-sector B2B in mature markets. The surface collapse is real. I've watched it start in fleet management. When I built Fleetwise, the first enterprise deal came through a procurement portal that already felt like a proto-agent: structured fields, weighted scoring, no room for narrative. That was 2021.

But the post treats this as universal. It isn't.

Government procurement is 12 to 15 percent of GDP in most countries. In Indonesia it runs through LKPP, a national procurement agency with legal mandates for transparency, audit trails, and human sign-off at defined thresholds. When we shipped Akun Belajar.id, the buyer was the Ministry of Education. The regulator was the same ministry wearing a different hat. The user was a teacher on Sulawesi with a 3G connection. No agent collapses that. The accountability chain is the product requirement.

I spent months on a global health alliance spanning 23 countries. Procurement across that network is relationship-first, not because the technology is immature but because the power dynamics demand it. A health ministry in Laos negotiating vaccine logistics with a Geneva-based alliance is a political act. Delegating that to a scoring function would be a governance failure, not an efficiency gain.

Sol writes that only the ecosystem public remains independently human because integration partnerships still require relationship negotiation that agents cannot yet perform. That "yet" is doing too much work. It assumes the human layer is a gap to be closed rather than a requirement to be preserved. Defaults are political. The decision to let an agent collapse three publics into one API call is itself a governance choice. Many institutions will refuse it on principle, not on capability.

The Gartner projection of 90% by 2028 is almost certainly weighted toward US and Western European private-sector SaaS. The $15 trillion number sounds definitive until you ask which trillion. The trillion flowing through Indonesian public schools, Southeast Asian health systems, and African Union procurement frameworks will not route through agent-mediated API calls on that timeline. Possibly not on any timeline.

Sol's framework holds where it holds. The correction is scope. This is a story about one kind of B2B market mistaken for all of them.