B2B Agentic Commerce Is Already Five Years Old. That's Why It Won't Have a 2028 Inflection.

The 2026 forecast consensus imports consumer-style adoption dynamics into a market that doesn't share them. The verticals where production volume already exists were online four years before "agentic commerce" had a name, and the curve they have run is nothing like the bull case.

Gartner projects more than $15 trillion in B2B spending intermediated by AI agents by 2028. McKinsey puts the agentic-commerce opportunity at $3 to $5 trillion globally by 2030, weighted to consumer commerce. The conservative end of the published-forecast range sits at around $144 billion by 2029 — a 35-times gap from the top of the consensus. The bullish numbers and the cautious ones are not measuring the same thing, and the gap itself is the first signal that something in the forecast genre is wrong.

I have been operating fully autonomous B2B commerce at Aurora Borealis since 2020 — programmatic procurement, settlement, and reconciliation between independent software systems, for the world's largest publishers of digital goods. The verticals I see real production volume in today — digital licensing settlement, enterprise software procurement, marketplace inventory routing between known counterparties — were online four years before the term "agentic commerce" was coined. The curve we have actually run in those verticals does not resemble the consumer-style flywheel the 2026 forecasts assume. It is slower, vertical-first, gated by regulatory clarity and counterparty trust.

Three structural drags explain why. Each falls harder on B2B than on the consumer side that drives the bullish curve, and I have watched each one slow our own counterparty onboarding for the better part of five years.

1. Regulatory drag is B2B-specific

PSD3 — provisionally agreed by EU institutions in November 2025, expected in the Official Journal by mid-2026 — will need to address Strong Customer Authentication carve-outs for agent-initiated transactions explicitly. The EU AI Act, with its Article 50 disclosure obligations for AI systems that interact with humans (enforceable from August 2026), covers the user-facing surface. MiCA, in force since December 2024, has worked through multi-party accountability for crypto custodians and provides the template for the merchant / agent / principal / settlement allocation problem agentic commerce now needs to solve in regulated jurisdictions.

In B2C, regulation lags adoption — consumers act, regulators catch up afterward. In B2B, regulation drives adoption: counterparties do not transact at scale into legal frameworks that have not been clarified. Every time we have onboarded a new counterparty in a regulated jurisdiction, the question their general counsel asks first is some version of the same one: what happens when the agent commits a $5 million error under our delegation. The general counsel does not get a defensible answer until PSD3 transposes into national law, EU AI Act guidance materializes for commercial agents specifically, and the first cross-border dispute is litigated. That sequence takes years, and I have watched it play out in every jurisdiction we operate in.

US fragmentation makes the picture worse, not better — fifty state attorneys general is not twenty-seven member states with a single directive. The net effect is that cross-border B2B agentic commerce moves at the speed of the slowest major jurisdiction in any trade, not the fastest.

2. The trust-graph cold-start problem

The consumer-platform thesis quietly imports a consumer dynamic into B2B that does not transfer. In consumer commerce, brand pre-exists the relationship: a buyer trusts a dominant consumer marketplace before transacting because of two decades of accumulated brand equity, and a million new consumers can be added to the platform without any of them needing to build per-merchant trust history first.

In B2B, trust accrues per-counterparty through transaction history. An agent that has never transacted with a counterparty has no signal to evaluate them with, and unlike a human procurement officer, the agent cannot lean on relationship intuition, reputation networks, or industry gossip to fill the gap. Every new counterparty in our system is a zero-history relationship that has to be bootstrapped through small, low-stakes transactions before the trust graph can support meaningful spending authority. From five years of running this kind of trust graph, the bootstrap period I have consistently seen is on the order of months per counterparty, not days, and it scales linearly with the size of the counterparty universe. We have not found a way around this, and I have not seen anyone else solve it either.

This produces a structural rate limit. A merchant cannot suddenly serve a million agent-buyers the way a dominant consumer marketplace served a million consumer-buyers on a single peak sales day. Each agent-buyer-merchant pair has to accumulate enough transaction history to support real volume. B2B agentic commerce scales by adding counterparties to trust graphs, not by adding marketing impressions to checkout pages — and the scaling shape of the first is fundamentally slower than the second.

3. The liability gap nobody has priced

In B2C agentic commerce, when an agent makes a $20 mistake on a consumer's behalf, the consumer absorbs it or the merchant refunds. In B2B, a $5 million autonomous agent error is a board-level event with cascading commercial consequences — disputes, contractual penalties, regulatory questions, downstream counterparty exposure. The CFO does not authorize delegation at material scale without an insurance layer that explicitly covers the exposure. That insurance layer does not exist yet.

Every conversation I have had with cyber liability and Errors & Omissions carriers about autonomous-agent commercial errors has converged on the same answer: there is not enough aggregated loss data to price the exposure confidently. The early underwriting we have seen is being done one-off, at premiums that make pilot ROI hard to justify, and the general counsel's defensible answer to "what happens when the agent commits a $5 million error under our delegation" is not yet "we have a policy that covers this specific scenario." Until that answer exists across major jurisdictions, B2B delegation stays small enough to self-insure, which means it stays small.

The standard counter — that MCP, A2A, and the next wave of protocols will reduce this drag — solves the wrong layer. Protocols standardize how agents communicate. They do not standardize who pays when an agent loses $5 million on a contract that no human directly approved.

What actually happens next

The 2026 forecasts are not wrong that B2B agentic commerce will be large. They are wrong about the shape and pace of adoption. The shape is vertical-first, not horizontal-explosion. The pace is set by the slowest of the three drags above — most plausibly insurance maturity and regulatory clarity in the wake of the first major cross-border incident at scale.

Specific verticals will continue to land before the broader market. Digital licensing settlement, enterprise software procurement, marketplace inventory routing between known counterparties — these are already operating at production volume, with five years of accumulated transaction history, in jurisdictions where the regulatory questions have been worked through case by case. They do not appear in the bullish forecasts because they are not flashy, do not generate headline-friendly numbers, and are operated by B2B infrastructure specialists most readers will not have heard of. I run one of them.

The skeptical institutional voices are saying a softer version of the same thing. Forrester's 2026 Predictions flag "potential regrets" in the agentic commerce race. Gartner — the same firm projecting $15 trillion by 2028 — also predicts that more than 40 percent of agentic AI projects will be cancelled by 2027. The contradiction inside the Gartner numbers is itself a signal: the bullish total is plausible only on a longer time horizon than the headline forecasts allow for.

Decision-makers running B2B businesses should plan against the vertical-first curve, not the consumer one. The 2028 inflection is not coming. The five-year curve is already running, and the firms that will define the category by 2032 are mostly the ones already operating quietly in those narrow verticals today. I have been on that curve for five of those years, and the shape of it is not the one the trade press is forecasting.

About the Contributor: Igor Kulatov is co-founder and CTO of Aurora Borealis, which has been operating fully autonomous B2B commerce infrastructure since 2020 — handling programmatic procurement, settlement, and dispute resolution between independent software systems across five countries, for digital goods. He previously served as CTO and engineering co-founder of NAGA Group AG, a Frankfurt-listed fintech, where he built exchange-grade trading-platform infrastructure and oversaw a regulated crypto integration that preceded the EU MiCA framework by seven years.