# When business agents talk to each other, who can agree to the deal?

Understand agent-to-agent commerce through a practical approval framework. Separate communication, identity, proposed terms and authority to commit money.

By Daltn editorial team. Published 2026-11-01.
Canonical: https://daltn.tech/feeds/agent-to-agent-b2b-commerce/

## Who can authorize a deal between business agents?

Business agents can exchange information and prepare proposed transactions, while communication protocols do not grant authority to make commitments. Define identities, permitted actions, terms and approval limits before allowing a system to accept prices or spend money. Begin with a bounded information request, verify the response and keep the agreement under accountable business control.

## Keep these in mind

- Communication capability and authority to commit are separate.
- Shared product definitions matter as much as fluent negotiation.
- Begin with information exchange before delegating purchases.

## A supplier conversation contains more than a price

A fictional café needs 200 cups by Friday. The supplier offers a cheaper item, but its volume, material and packaging differ.

A system optimizing only for the lowest quoted price may buy an unsuitable product. The business needs a definition of acceptable substitution.

Delivery terms, tax, minimum quantity, returns and payment terms also affect a deal. An agent should not infer permission to change them merely because it can exchange messages. The owner must define what may be explored and what may be accepted.

## Protocols make coordination possible, not automatically trustworthy

The [Agent2Agent specification](https://a2a-protocol.org/latest/specification/) defines concepts such as agent discovery, tasks, messages and authentication-related metadata. These give systems a shared way to communicate. They do not prove that a discovered agent represents the supplier you intended or that its offer is commercially acceptable.

Check business authority separately. Establish the organization represented, the account acting and the commitments that account may make.

A conversation that looks convincing is not enough. Retain identifiers and evidence of the accepted terms so a person can reconstruct the exchange.

## Use established document meanings

Structured business exchange predates language-model agents. [OASIS Universal Business Language](https://docs.oasis-open.org/ubl/UBL-2.4.html) defines reusable structures for documents including orders, responses and invoices. Such standards illustrate why stable fields and meanings matter alongside natural-language negotiation.

For the cups example, quantity should specify whether it means individual cups or cases. Delivery should identify a date and destination.

A unit price should include currency and the relevant tax or charge treatment. An AI-generated summary can explain these terms, but the accepted record needs the precise values.

## Set a permission boundary before the request

The following policy is an illustrative trial for a fictional café requesting a supplier quote. It keeps information gathering separate from the authority to agree a purchase.

Activity | Trial permission | Required check
--- | --- | ---
Ask availability | Allowed for a named product | Verified supplier endpoint
Request a quote | Allowed without commitment | Exact quantity, unit and destination
Suggest an alternative | Prepare for owner review | Material and capacity requirements
Accept terms | Owner approval required | Full final price and delivery terms
Initiate payment | Outside trial scope | Authorized financial process

## Expect stale offers and incomplete agreement

An offer can expire while the buyer considers it. Stock can change. One side may treat a reply as acceptance while the other treats it as a question.

Define states such as requested, quoted, awaiting approval, accepted and cancelled. Do not let an optimistic summary skip the final acceptance.

A duplicate request should not become two orders. A timeout should not be treated as rejection or acceptance without checking.

Keep a stable request or order identifier and an explicit final status. If the systems disagree, give people an accessible way to inspect the exact terms and stop further action.

## A sensible near-term progression

Start by asking a verified supplier for an availability statement. Then try preparing a comparison of quotes for human review.

Before permitting commitments, define limits and verify identities. Test recovery procedures and confirm that agreed terms are recorded reliably.

A practical pilot can stop at preparing an offer comparison for a person to decide. Measure whether it preserved product requirements, explained differences and retained the supplier’s actual terms.

Record questions that required judgment, such as a substitute material or changed payment condition. Those exceptions show which decisions are ready for a rule and which still benefit from a purchasing specialist.

1. Choose one supplier and a read-only information request.
1. Define exact products, units and required terms.
1. Verify both system identity and business authority.
1. Require review before any commitment.
1. Preserve the final agreement and handle duplicate or failed exchanges.

## Common questions

### What should a business agent be allowed to negotiate first?

Begin with a bounded information request, such as availability or a nonbinding quote for a precisely identified product. State quantity, unit, destination and required terms. Keep purchase acceptance and payment behind the business’s explicit authority and verification process.

### What must be retained after systems exchange an offer?

Keep the verified parties, exact product, quantity, currency, delivery and payment terms, plus the accepted status and stable request identifier. Record revisions and unresolved conditions. A fluent summary should not replace the precise agreement the people responsible need to inspect.

## Sources

- [Agent2Agent Protocol specification](https://a2a-protocol.org/latest/specification/): Defines agent discovery, task exchange and related protocol objects.
- [OASIS Universal Business Language 2.4](https://docs.oasis-open.org/ubl/UBL-2.4.html): Defines structured business documents and order-response semantics.