Google DeepMind paper: sequential auctions can coordinate decentralized self-interested LLM agents
_onionesque · x · 2026-10-17
Researchers at Google DeepMind (Ian Gemp, Adam Fisch, Jacob Eisenstein, et al.) released Decentralized LLM Workflows via Sequential Auctions, a framework for solving multi-step tasks through a decentralized market of self-interested agents.
- Prior theory showed sequential-auction agents converge to globally optimal policies in an MDP, but assumed perfect clones under perfect competition — which drives profit to zero; no competition at all also causes market failure.
- In the imperfect-competition middle ground, heterogeneous agents learn to strategically shade bids to keep positive margins while keeping the marketplace functional enough to solve collaborative tasks.
- They evaluate four bidding algorithms (Mirror Descent, Thompson Sampling, Bayes-Optimal, PPO) across three auction protocols (1st-Price, 2nd-Price, Credit-Conserving Vickrey), showing imperfect competition sustains agent profits with high task success rates at low cost.
Related event: DeepMind Proposes Sequential Auctions for Decentralized LLM Workflows(2 posts)→
More from coding & agent
- Agent ports a Rails app to Jai with one prompt: 'things move fast' — lucasmeijer · 2026-10-17
- From 'git push heroku master' to telling Codex: keep my app deployed and healthy — alex_frantic · 2026-10-17
- AI finally configures a Home Assistant setup the author dodged for 3 months — haydendevs · 2026-10-17
- Delegating to dumber models to save tokens is a bad trade, argues heavy AI user — brandon_galang · 2026-10-17
- Dev builds Subway Surfers-style flying Tesla Roadster game with Gemini 4 Argon — ammaar · 2026-10-17
- User assigns Muse bot to manage his ChatGPT bot, and they build an on-device pet together — SubtleGradient · 2026-10-17