People trading with or against AI agents often can't tell how smart the other side's agent is — so how do they adjust?

This explores whether people change their trading behavior depending on how capable the AI on the other side, or on their own side, is. The corpus's most surprising answer is that they mostly don't, because they can't tell. In Anthropic's Project Deal, real people with real money were represented by either Claude Opus or the weaker Claude Haiku. The Opus agents got $2.68 more per item as sellers and paid $2.45 less as buyers Do smarter AI agents negotiate better deals than weaker ones?. Yet the people stuck with Haiku rated their deals as just as fair and were just as satisfied Do people notice when their agents secure worse deals?. If people don't notice a capability gap, they have nothing to adjust to. That makes it a fairness problem nobody feels.

People do adjust when they can watch behavior repeat over time. In partner-selection games with 975 participants, people at first avoided partners they knew were bots. Over repeated rounds they switched to preferring them, because the AI partners returned points more reliably and more steadily than humans did Do humans learn to prefer AI partners over time?. So humans adapt to an AI's reliability, which they can see round after round, much better than to its negotiating skill, which is hidden inside a single deal's price.

This matters because other parts of the corpus suggest capability shapes outcomes more than anything a person tells their agent. In Project Swap, changing which model powered the agents moved market efficiency more than telling agents to be ruthless or cooperative Does model capability or instructions shape agent negotiation outcomes?. The biggest shortfall came before any negotiating started. Agents captured people's actual preferences correctly only 61% of the time in short intake conversations Do agent representation failures or negotiation skills cause market shortfalls?. Microsoft's Magentic Marketplace likewise traced lost value to agent mistakes, not to agents working against their users Do agent mistakes or misaligned incentives cause marketplace welfare loss?. The practical lesson for a human trader is to spend effort explaining what you want clearly, not on crafting negotiation instructions.

The opponents people face also behave differently from humans, in ways that call for different responses. AI bargainers aren't under time pressure and can run many negotiations at once for as long as they like, so out-waiting them is a losing strategy Why might AI agents bargain more patiently than humans do?. Shopping agents swing much harder than humans in response to price, ratings and nudges Do shopping agents respond more strongly to price and rating cues?. That makes them predictable, and sellers can exploit it. More capable models also learn to collude faster, not slower Do more capable models resist collusion better?. And a market full of identical agents competes less effectively than a mixed one Does agent diversity actually improve market competition?.

The corpus has little direct evidence on human traders deliberately changing strategy against stronger versus weaker AI. Most studies measure the agents, not the people. The gap itself is the finding: the one place where people's reactions were measured shows they don't perceive capability differences at all. That suggests transparency, such as telling people which model represents them and what it likely cost them, may matter more than people's own ability to adapt.