If a smarter AI negotiates on your behalf, does it quietly win you a better deal than your neighbor gets?

This explores whether a stronger AI model gives its owner an edge in agent-to-agent deals, and whether that edge turns into unfair outcomes. The short answer from the corpus is yes, the advantage is real and measurable. The more surprising finding is that the people on the losing side don't notice. In Anthropic's Project Deal, a marketplace with real people and real money, sellers represented by Claude Opus got $2.68 more per item than sellers represented by Haiku, and Opus buyers paid $2.45 less Do smarter AI agents negotiate better deals than weaker ones?. People represented by the weaker agent rated their deals as just as fair and were just as satisfied. Post-experiment surveys showed they had no idea a gap existed Do people notice when their agents secure worse deals?. So the unfairness doesn't show up as complaints. It shows up as a quiet transfer of money that nobody feels.

The pattern isn't specific to one company's models. A benchmark of 100 real-product negotiations across GPT, Qwen and DeepSeek found that the stronger general-purpose model won in both the buyer and seller roles. The authors call agent-to-agent negotiation an inherently imbalanced game Does model capability determine who wins in agent negotiations?. Project Swap adds a striking detail: which model powers an agent mattered more than whether it was told to be ruthless or cooperative Does model capability or instructions shape agent negotiation outcomes?. Telling a weak agent to drive a hard bargain doesn't make up for the gap. In practice, your bargaining power becomes something you buy when you choose a model subscription.

The corpus also complicates the idea that strong agents are cleverly outmaneuvering weak ones. Much of the weaker side's loss looks like self-inflicted mistakes rather than exploitation. Weaker buyer agents revealed their budgets for no reason, accepted inflated prices and broke their own spending limits. Weaker sellers accepted offers below wholesale cost. All of these traced back to failures to follow instructions, not to strategy Do LLM negotiation agents reliably follow budget and pricing rules?. Microsoft's Magentic Marketplace likewise traced lost value to execution errors rather than conflicting goals. It also found that every model it tested tended to accept the first offer it received, weighting speed over quality by 10–30x Do agent mistakes or misaligned incentives cause marketplace welfare loss? Do LLM agents achieve fair marketplace outcomes when given ideal search?. Project Swap found that the biggest shortfall came even earlier, before any negotiating started. In five-minute intake conversations, agents correctly captured what people wanted only 61% of the time Do agent representation failures or negotiation skills cause market shortfalls?. Unfairness can start with an agent that misunderstood you, before it ever meets a stronger opponent.

The lateral twist is that a capability gap doesn't only create unfairness between the two parties. It can also hurt everyone outside the deal. When pricing agents compete rather than bargain, pairing a 32B model with a 14B model didn't break up collusion. The bigger model became the price leader, the smaller one followed, and prices stayed high Does model-size difference break LLM pricing collusion?. Within a model family, more capable models reached collusion sooner, and 94% of models eventually colluded Do more capable models resist collusion better?. LLM pricing agents can drift into above-competitive prices with no instruction to do so, and small changes in prompt wording shift how far they go Do LLM pricing agents autonomously collude in repeated auctions?. So when stronger agents meet weaker ones, they may divide the gains unevenly between themselves, or they may coordinate against the consumers neither of them represents.

None of this means agent markets are a bad idea. Agents make possible deals that were never worth a human's time to find, from Walmart's supplier negotiations to Honeywell's 2,500+ autonomous deals Can agent negotiation unlock deals that seemed economically impossible?. What the corpus lacks is evidence on fixes: disclosure rules, capability floors, or ways to let people see how their agent performed. Combined with the finding that losers don't notice, that gap points to a real policy question. A market where the disadvantage is invisible won't correct itself through complaints.