Can agent negotiation unlock deals that seemed economically impossible?

Sangeet Paul Choudary argues that the larger opportunity in the "emerging agent economy" is not agents making existing transactions more efficient, but agents "making transactions possible that would otherwise never happen." He frames this against the common framing of agent commerce, which "focuses on helping buyers find better offers and negotiate better prices" within offers sellers have already made available. Choudary's claim is that once sellers are also represented by agents, "they are no longer limited to negotiating over what has already been offered" — both sides "can begin to construct offers that didn't exist before the interaction." As evidence of the pattern already operating, he cites Walmart using AI to negotiate individually with suppliers over "prices, payment terms, and other commercial conditions," Honeywell completing "more than 2,500 autonomous negotiations covering over $500 million in spending," and Anthropic's Project Deal, where 69 employees represented by AI agents concluded "186 real agreements involving goods worth more than $4,000" without principals managing the process.

The mechanism he gives is cost, not capability: many valuable transactions "never happen — not because buyers and sellers are fundamentally incompatible, but because discovering and constructing an agreement is too expensive." His manufacturer example makes the point concrete — a buyer needs 5,000 units with requirements no single supplier meets, and the deal is abandoned even though a logistics provider and a certification partner could, in combination, have made it work; the market simply never searches for that combination because doing so isn't economical for a human-run process. Agents, by lowering the cost of "exploring alternatives, negotiating trade-offs, and coordinating agreements," can make such combinatorial deals viable. He adds a further requirement for this to generalize: agents must understand not just what a buyer or seller is literally asking for or currently offering, but what each is "trying to achieve" or "capable of providing" — a job seeker open to a different role, a hotel able to recombine services — which in turn requires agents to be connected to the operational capabilities and commercial decisions of the organizations they represent and able to interact across organizational boundaries.

This cuts against the market lens in Why do AI-delegated firms stop exploring new business models?, where full AI delegation of strategy settles into a self-confirming equilibrium that makes novel business-model implementations rare because nobody has incentive to search outside the agent's existing frame; Choudary's argument is the reverse direction — bilateral agent representation is what expands the frame of discoverable deals, at least in bounded bilateral negotiation rather than firm-wide strategy. It also sits beside, rather than against, Why does manual search fail as product choices become complex?: that model treats agentic commerce as solving a buyer's search-and-match problem within an existing catalog of seller offers, while Choudary's claim is that agent representation on the seller's side removes the fixed-catalog assumption itself, letting the catalog be constructed during negotiation.

The excerpt is an essay, not a study — Choudary offers no counterfactual showing the cited deals genuinely would not have happened without agents, and the Walmart, Honeywell, and Project Deal examples are illustrative anecdotes rather than a comparison group. Project Deal, by his own account, was small (69 employees, under $4,000 in goods), so its evidence for generalizing to organization-to-organization commerce is limited. The argument also assumes agents can be reliably connected to the operational capabilities and commercial decisions of the firms they represent — a precondition the essay states as a requirement rather than something it shows any of its cited examples has achieved end to end.