Where does an AI assistant's real value actually come from?

The excerpt reports that in its simulated rental market, "the simulated consumer reads the fine print on most days — hundreds of times field readership — and still pays, while an assistant holding a concrete, approved plan completes the cancellation almost always." The paper names this an "acting-not-reading account of assistant value": the channel through which an assistant helps a consumer is not better information but follow-through on an action the consumer has already approved and still fails to execute.

The excerpt separates the surviving value "down to its two components — fee avoidance and task completion — of which only the first is contestable by sellers." Fee avoidance is large at first ("on day one, closer to three-quarters" of the value) but "self-liquidating," since sellers reprice headline rates to recapture it. Task completion, by contrast, persisted "in every configuration we ran," measured "within the society" at "3.75 ± 0.64 USD per renter-day," and was "dented only where fee-free selling rescues the unshocked side's rentals as well." The paper's own framing of the split: "Products sold on fee savings price their own obsolescence; products built on execution reliability sell what the market left standing."

This decomposition sits alongside Do sellers recapture savings when AI assistants help buyers?, which covers the fee-avoidance half of this same split — the half sellers successfully reprice away. It also qualifies Why do consumers delay adopting agentic search?, since both ask what part of an assistant's benefit survives strategic response, though this excerpt locates the durable part in task completion rather than in search itself. And it bears on Can EU consumer law protect AI agents as algorithmic buyers?: both treat an assistant's authorized action, not the information it surfaces, as the operative unit worth regulating or valuing.

The excerpt cautions this is "an LLM's follow-through failure priced at the taxi fallback," and "though it responds to fees as human non-booking would, its level need not transfer to people." It does not claim the 3.75 USD figure, or the near-complete cancellation rate, would hold for human renters or other contract designs — only that within this construct, the acting component of assistant value proved structurally harder for sellers to compete away than the reading or informing component would have been.