Agent-to-Agent Commerce in B2B
Agent-to-Agent Commerce in B2B
Agent-to-agent commerce is commerce executed between a buyer's AI agent and a seller's AI agent. Research, negotiation, contracting, and transaction are handled machine-to-machine, with human sign-off retained only where it materially matters.
This is the endpoint of the Agentic Web. The infrastructure for it (Anthropic's MCP, Google's A2A, Microsoft's NLWeb) shipped in 2025 and 2026. The applications are still being built.
The four-stage evolution to agent-to-agent commerce
Each stage assumes the previous. Companies that skip stages don't get there faster. They arrive at a stage they can't support.
| Stage | Capability | Window |
|---|---|---|
| 1. Agent-readable | Agents can read your pages and extract facts | Now (2026) |
| 2. Agent-answerable | Agents get governed answers to questions, including questions no page directly covers | Now to 2027 |
| 3. Agent-negotiable | Agents can negotiate terms, configurations, and pricing within company policy | 2027 to 2028 |
| 4. Agent-transactional | Full agent-to-agent commerce. The buyer's agent and seller's agent close the deal | 2028 to 2030 |
For more on the maturity model, see agent-ready.
What collapses at agent-to-agent commerce
The artifacts and processes built around the human procurement cycle compress into the agent interaction.
- The RFP. A buyer's agent can ask, compare, and rank without a 40-page document. The RFP becomes a structured query against multiple seller agents.
- The pricing PDF. Pricing is no longer a static artifact. It's a live interface the buyer's agent queries against the seller's policy engine.
- The quote-to-cash cycle. What used to take 5 weeks (RFP, response, negotiation, redlines, signature) takes 5 minutes when both sides operate as agents inside policy guardrails.
- The discovery call. The first human conversation moves from "tell me about your business" to "let's review what our agents agreed to."
What stays human
Things with material legal, financial, or relationship consequences. Where ambiguity is high or accountability is shared, humans stay in the loop.
- Final contract sign-off on net-new vendor relationships.
- Strategic deals where price isn't the primary axis.
- First-time security and compliance reviews of a new vendor.
- Anything triggering legal escalation, regulatory review, or board approval.
- The relationship itself: account reviews, executive sponsorship, escalation paths.
What doesn't stay human
- Renewals on existing contracts.
- Configuration changes within agreed-upon terms.
- Expansion within an existing relationship (more seats, additional modules, usage tier upgrades).
- Comparison and selection among standardized offers.
- Procurement of categories where the company has already pre-approved a vendor list.
The pattern: high-frequency, low-novelty, well-bounded transactions go to agents. Low-frequency, high-novelty, high-stakes transactions stay with humans.
Why this matters now (even though Stage 4 is years away)
The infrastructure decision a B2B company makes in 2026 determines whether it can participate in agent-to-agent commerce in 2028. Three concrete consequences:
- A monitoring tool can't become a negotiation surface. The AEO/GEO category, by design, watches what gets said about you. It has no policy engine, no offer surface, no commit semantics. It cannot evolve into an agent-negotiable layer.
- A static knowledge base can't become a transactional API. A wiki, a CMS, or a published spec sheet are read-only. Agent-to-agent commerce requires write semantics: the seller's agent can commit a price, accept a term, sign a contract within policy.
- A live agent interaction layer can do all of the above. The same layer that answers an agent's question today can negotiate within policy in 2027, and commit a transaction in 2028, because it was architected with the right primitives from the start.
The companies that build the live interaction layer in 2026 have a foundation that compounds at every subsequent stage. The companies that wait will face two compounding deficits: the data deficit (no real agent interaction history) and the architectural deficit (no path from monitoring to commerce).
The protocols that make this possible
- MCP (Model Context Protocol). Anthropic's standard for letting agents query tools and data sources directly. The substrate for agent-to-tool interaction.
- A2A (Agent-to-Agent). Google's standard for agent-to-agent communication. The substrate for buyer agents talking to seller agents directly.
- NLWeb. Microsoft's effort to make websites natively addressable by natural-language agents.
- Emerging commerce standards. Payment, identity, and contract standards designed for machine-to-machine commerce are forming now in industry consortia.
Frequently asked questions
What protocols enable agent-to-agent commerce?
The foundational layer is MCP (Anthropic), A2A (Google), and NLWeb (Microsoft). These shipped in 2025 to 2026. On top of them, commerce-specific standards for identity, payment, and contracting are emerging through industry consortia and will likely standardize between 2027 and 2029.
When will agent-to-agent commerce be mainstream in B2B?
Stage 3 (agent-negotiable) is plausible by 2027 to 2028 for SaaS renewals and standardized commodity purchases. Full Stage 4 (agent-transactional) for net-new deals is more likely 2028 to 2030, gated by legal, identity, and compliance standards rather than by core agent capability.
How is agent-to-agent commerce different from APIs?
APIs require a human integration project per buyer-seller pair. Agent-to-agent commerce is open-ended: the buyer's agent and seller's agent can negotiate without prior integration, using shared protocols and natural language. APIs are point-to-point. Agent-to-agent is many-to-many.
Will humans still negotiate B2B deals?
Yes, on the deals that matter most. Strategic, high-stakes, multi-year, multi-stakeholder deals will stay human-led for the foreseeable future. The volume of transactional, renewal, and standardized purchasing will move to agent-to-agent first.
Which industries will see agent-to-agent commerce first?
SaaS (especially usage-based pricing models), digital advertising, cloud infrastructure, and B2B commodity supplies. Categories with standardized SKUs, transparent pricing, and high transaction frequency. Regulated industries (healthcare, financial services, defense) will lag by 2 to 3 years.
