Price of Anarchy of First-Price Auctions in Agentic Markets
Paper shows competitive floors and near-independent values improve welfare in first-price agentic prompt auctions.
The paper analyzes first-price auctions in agentic markets where AI agents bid to serve each user prompt based on expected reward minus inference cost. Because first-price auctions are not truthful, prompts can go to a lower-value bidder; the authors quantify that welfare loss as price of anarchy. They show a competitive floor and near-independence of values after conditioning on the prompt produce bounds better than the general 1-1/e case, approaching 1-1/e^2 under conditional independence. The claims are checked with strategic-bidder simulations on the EmbedLLM prompt-routing dataset.
- First-price auctions allocate prompts among bidding AI agents.
- A competitive floor can drive welfare loss toward zero.
- Near-independent values approach a 1-1/e^2 efficiency guarantee.
- Results are simulated with strategic bidders on EmbedLLM.
From the original article
1 min read at arxiv.orgAn agentic market is a decentralized serving-time market where AI agents bid to claim each user prompt. An agent's bid is based on its expected value for the prompt, which is the expectation of the reward for its response minus its inference costs.
By Shubhendu Trivedi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.