Agent-to-Agent Commerce: The Quiet Revolution in B2B Transactions

Brian Rozelle

Digital Transformation and AI Strategy Consultant

Founder, Rozelle.ai

Published October 26, 2026

Your next big sale might not be negotiated by a person. It might be negotiated by your system talking to their system. And your system might not even know the conversation is happening.

Deloitte’s 2026 B2B commerce research, surveying over 1,000 US suppliers and buyers, reveals a market that is ambitious but unevenly prepared. Forty percent of B2B buyers already use agentic AI in purchasing. Only twenty-four percent of suppliers use agents in sales. Buyers are six times more likely than suppliers to describe B2B processes as mostly manual. That gap is a competitive blind spot.

If you sell to other businesses, your customer’s procurement agent may already be evaluating you. The question is whether your business can talk back.

What Is Agent-to-Agent Commerce?#

Agent-to-agent commerce is the practice where AI agents representing buyers negotiate, evaluate, and transact directly with AI agents representing sellers, with minimal or no human involvement in the transaction itself. It sits at the far end of a continuum that most businesses are only halfway through.

Deloitte frames this as a four-stage progression:

Stage 1: Agent-assisted. AI helps with discrete tasks like supplier discovery or contract analysis. The human is the primary operator.

Stage 2: Semi-autonomous. Agents coordinate multiple tasks internally, running RFQs, scoring proposals, and drafting contracts. Humans define strategy and manage exceptions.

Stage 3: Outcome-driven. Self-optimizing agents pursue business goals across procurement. Humans set objectives and intervene in complex cases.

Stage 4: Fully autonomous. Buyer and seller agents transact directly across organizations. Humans set policies and guardrails, but the machines handle the deal.

Most businesses today hover between Stage 1 and Stage 2. But the shift to Stage 3 and 4 is accelerating. A hospital buying cardiac stents might start with an agent-assisted search, move to semi-autonomous proposal scoring, and within two years find its procurement agent negotiating directly with a manufacturer’s sales agent on price, delivery, and terms.

The human does not disappear. The human’s role shifts from operator to overseer.

The Infrastructure Is Arriving (But Commerce Is Already Happening)#

On October 6, 2026, Meta and Sierra announced the Personal Agent Protocol. It is an open, OAuth-based standard that governs how personal AI agents authenticate and interact with businesses. Partners include Stripe, Shopify, Walmart, Genesys, and Rocket.

Notably absent from the coalition are Amazon, OpenAI, and Anthropic. Bret Taylor, Sierra co-founder and OpenAI Chairman, expects them to join eventually. But the split creates friction that may slow universal adoption in the short term.

Here is what the headlines often miss. The protocol is infrastructure. Commerce is already flowing. Shopify reported that orders from AI-powered searches grew nearly thirteenfold in the first quarter of 2026. New buyers placing orders through AI channels did so at nearly twice the rate of other channels. The agents are not waiting for permission.

For a small business, the question is not whether to adopt agent-to-agent commerce. The question is whether your systems are discoverable when an agent comes looking.

The Scale of What’s Coming#

The numbers are hard to ignore. Gartner forecasts that AI agents will intermediate more than fifteen trillion dollars in B2B purchases within three years. Ninety percent of B2B purchases are on track to be AI-agent intermediated by 2028.

The AI agent market is estimated at thirty-five billion dollars by the end of the decade, potentially forty-five billion if governance keeps pace.

On the supplier side, the incentives are stark. Suppliers lose an estimated thirteen percent of sales bids to negative buyer experiences. Positive buyer experiences drive an estimated thirty-six percent revenue uplift. Buyers spend nearly thirty percent more with suppliers that deliver positive experiences.

Ninety-two percent of buyers currently using EDI, the forty-year-old standard for electronic data interchange, plan to shift to more agent-friendly channels. Nearly ninety percent of B2B suppliers are upgrading or preparing to upgrade their ERP systems. That creates a window. The businesses that build agent-ready infrastructure now will be the ones that win the evaluations later.

The Risks: When Machines Make Deals#

As agents move from assistance to autonomous negotiation, the risks multiply. They fall into two categories: operational and strategic.

Operational risks are the immediate concerns. How do you verify that a counterparty agent is who it claims to be? Emerging frameworks like Know Your Agent, modeled on Know Your Customer banking standards, are beginning to address identity verification. But standards are still patchy. If an agent agrees to terms, who holds the record? Who is liable if the agent misunderstood?

Strategic risks run deeper. Agents optimize for price and speed. They can miss value-adds, relationship factors, and long-term strategic fit that human negotiators catch. A procurement agent might drive a supplier’s margin to zero without realizing that supplier is the only one capable of fulfilling rush orders during peak season. Negotiation patterns can also reveal competitive intelligence. If your agent always pushes for thirty-day payment terms, your counterparty knows your cash flow pressure.

The IMF’s research on agentic AI and payments emphasizes human-in-the-loop supervision as a core safeguard for high-risk or high-value transactions. Thresholds requiring human approval, supervisory dashboards, and manual override mechanisms before legal finality are prerequisites for responsible deployment.

Human Checkpoints: Keeping Control Without Slowing Down#

The best agent-to-agent commerce implementations do not eliminate humans. They place humans precisely where judgment adds value.

Strategic supplier selection is a human job. An AI can score proposals on price, delivery, and compliance. A human evaluates soft factors, long-term fit, and alignment with company values.

Final contract negotiation and sign-off requires human insight. AI drafts and analyzes terms well. But a human balances legal risk, business priorities, and market dynamics in ways that agents still cannot.

High-value transactions should trigger automatic human approval. The IMF recommends explicit thresholds. Anything above a defined dollar amount, or involving new suppliers, or carrying regulatory exposure, stops at a human desk before it becomes binding.

Edge cases and exceptions are where automation breaks. High-volume, low-value tasks can flow freely. Exceptions require human intervention.

The practical framework is simple. Define transaction thresholds. Build supervisory dashboards for real-time agent decision monitoring. Create manual override mechanisms before legal finality. Maintain transparent AI decision logs. Designate human drop-in points at key workflow stages.

These are not enterprise-only practices. A five-person supplier with a clear approval threshold and a shared decision log is more prepared for agentic commerce than a hundred-person firm flying blind.

What to Do Before Your Customer’s Agent Calls#

You do not need a custom AI agent or a six-figure systems integration to prepare. You need clarity and a few afternoons of work.

Audit your digital presence. Can an AI agent, reading your website without human guidance, figure out what you sell, how much it costs, and how to buy it? If not, fix that first. Agents read text the same way search engines do. Clear descriptions, transparent pricing, and easy-to-find policies are not just good SEO. They are agent readiness.

Review your booking or payment platform. Do you use Shopify, Stripe, or another platform with API access? Much of the groundwork for agent interaction may already be in place. Check whether your systems support OAuth or clean API standards.

Document your terms and policies. Agents evaluating suppliers will read return policies, cancellation terms, and shipping details before they engage. Make these obvious and machine-readable.

Set internal thresholds. Decide now what an agent can do without human approval and what requires a person. A simple rule like “any order over five thousand dollars needs a human sign-off” prevents surprises.

Start with discoverability. The Personal Agent Protocol, Google’s A2A, and Anthropic’s MCP all converge on one idea: agents need to find you before they can buy from you. A clean, fast, well-structured website is the first and most important step.

The businesses that thrive in agent-to-agent commerce will not be those with the best human salespeople. They will be those whose systems can negotiate intelligently with their customers’ agents. And the real risk is not being replaced by AI. It is being invisible to it.

“Want the tools to match the vision?” Explore our digital products at Rozelle.ai ↗ - built for business owners who want to lead with AI, not follow.