When Agents Start Negotiating

 Earlier this year, on the same day, two announcements landed that together signal where agentic AI is heading, and what it will ask of the organisations adopting it. 

1. Visa published its When AI Becomes the Customer report, introducing what it calls B2AI: business-to-AI commerce. 53% of surveyed US businesses said they would allow AI agents to negotiate directly with other AI agents. 71% said they are willing to optimise products and offers specifically for AI agents. This is a global payments company is no longer treating agentic commerce as a thought experiment. It is positioning trust, override controls and machine-to-machine negotiation as infrastructure.
   
   
2. Microsoft released its Agent Governance Toolkit, an open-source project built to secure autonomous agents at runtime rather than at build time. The toolkit addresses the ten risks in the OWASP Agentic AI Top 10, includes identity controls, policy enforcement, circuit breakers and a kill switch for emergency termination. The market signal is clear: The question has moved from "how do we build agents" to "how do we safely govern fleets of them once they act inside live systems."

For organisations, the translation is uncomfortable but necessary. In your organisation today, who is accountable for what an agent does on your behalf? If an agent negotiates a poor contract at 2am on a Sunday, authorises a refund it should not have or misrepresents your brand in a customer interaction, who owns that outcome?

The tools are new, the question is not.

Whoever deploys an agent still owns what it does, which means accountability does not fade as autonomy rises and it concentrates.

That is a governance question, not a technical one, and it lands the same way in every sector. Payments face it first, because money moves fastest. Regulated industries face it hardest, because in life sciences an unowned decision is an audit finding waiting to happen. But the discipline is identical wherever you sit. Every agent acting on your behalf needs a named owner, a defined boundary, and a record that can show both.

Agentic commerce has arrived. The organisations ready for it are not the ones with the most agents. They are the ones who can still answer a simple question, at 2am on a Sunday: who is accountable, and can we prove it?

About the author: Siobhán O'Leary, AIGP, is an Applied AI Advisor and co-founder of The Institute of Applied AI, helping organisations build AI capability that is grounded in literacy, governance, and practical adoption. She publishes AI in Motion, a weekly newsletter for leaders navigating AI beyond the headlines.

Subscribe to AI in Motion on LinkedIn →