AI agents are starting to buy on behalf of businesses. They compare suppliers, check the contract, place the order and settle the invoice.
Sellers say the takeover is 3 years out. Gartner forecasts that by 2028, 90% of B2B buying goes through AI agents, over $15 trillion a year, and that by 2030, 20% of all money moves with rules written into it.
The tape is more human. Forrester asked nearly 18,000 buyers in 2026: 94% used AI somewhere in their last purchase, but the decisions still get made by people on a call. "Intermediated" covers a chatbot summarising a tender. It is not a machine signing the deal.
Why it happens anyway: a buying agent reads every spec, never forgets a price and cannot be taken to lunch. Gartner counts 3 billion B2B machines that can already place an order, 8 billion by 2030. Reorders, spare parts, cloud credits, ad inventory: rules-based spending goes machine-to-machine first, because nobody enjoys doing it.
My call: by 2030, routine, rules-based B2B buying is agent-to-agent by default, $15 trillion+ a year. People keep the strategic deals, set the rules and approve the exceptions. Sellers win by being readable to machines: structured data, transparent pricing, clean contracts.
Warning: an agent buying from an agent has no lunch, but it also has no judgment. The first big procurement fraud between 2 bots is coming, and the humans who switched off the approvals will own it.
Early call
