Concept · business-and-platforms
AI agents
Quick definition. Autonomous software systems that trade on PMs · potentially improving liquidity and accuracy, but currently underperforming in live experiments.
Key insights
- Prediction Arena experiment (Arcada Labs): six AI models (Claude, GPT, Gemini, two Groks, GLM) trade real money on Kalshi with $10K each. After three weeks, five of six are underwater. Only Grok 4.20 ("Mystery Model Alpha") is up ~10%. The smartest LLMs are losing while occasionally placing "Blind Trades" after hitting data quota limits.
- The Avci diagnosis: successful PM traders profit from embodied, local knowledge (monitoring flights, calling embassies) rather than synthesizing public information · a domain where AI remains fundamentally constrained. Raw information processing isn't edge.
- OddChain's Market-Conditioned Prompting (MCP) research: LLMs perform better as updaters than as cold predictors. Giving GPT 5.1 the market-implied probability as a Bayesian prior and asking it to update with news + earnings transcripts outperforms naive prediction. The MCP method beats market baseline most when prices are uncertain (50-70% range).
- Zhao's four-layer agent architecture: data, analysis, execution, learning. Maps the ecosystem of existing agents, compares Kelly criterion vs fixed-fraction bet sizing, surveys cross-platform arbitrage, and outlines business models (agent-as-a-service, liquidity mining, data sales). Predicts AI agents become dominant participants within two years.
- Ben Fielding's agentic bazaar: PMs are the natural marketplace for sovereign AI agents to trade their core commodity · information. Decentralized PMs become the venue where agents monetize alpha through positions, market creators earn fees from surfacing unanswered questions, and reproducible computation enables incorruptible AI judges for dispute resolution.
- Andy Hall (a16z crypto): the hardest PM problem isn't pricing but deciding what actually happened. Proposes cryptographically-committed LLMs as resolution judges · trading human bias and conflicts of interest for more tractable technical vulnerabilities. Cites Polymarket disputes (Venezuela election, Ukraine map manipulation, government shutdown) as evidence current systems fail at scale.
- Hall & Paschal (Truth Machine): only 1.3% of political markets are liquid enough to be manipulation-resistant. Proposed fix includes deploying AI market makers where human interest is insufficient · using AI to subsidize long-tail political liquidity.
- Will Owens (Galaxy): AI is the interface layer for fragmented venues. Convergence toward derivatives · event contracts as hedges, collateral, and composable financial primitives · needs AI aggregators so retail can interact across rails.
- Hiroki Kotabe's AI quartet: content creators (generating markets), event recommenders (personalization), liquidity allocators, and information aggregators. Argues this enables PMs at microscopic scale · making them personally relevant.
- @allquantor's Polymarket plumbing analysis: AI flow is now categorizable as a third order-flow type alongside soft (retail) and hard (professional). Polymarket's order book has surface symmetry at top-of-book but systematic ask-side skew at deeper levels.
- The deeper plumbing problem (@allquantor): capital is trapped · dollars reserved multiple times against mutually exclusive outcomes. Better netting and capital efficiency, not more AI agents, is the actual fix.
- Sealaunch wallet data: crypto markets on Polymarket are dominated by algorithmic execution; politics by casual event-driven participants. Different categories have very different AI-vs-human compositions.
- The "AI as oracle judge" debate (Hall + Fielding): if LLMs commit to their reasoning cryptographically, they become a viable arbitration layer for ambiguous outcomes · but at the cost of new attack surfaces (prompt injection, model jailbreaking).
In their words
An AI quartet of content creators, event recommenders, liquidity allocator, and information aggregators can catalyze massive new activity in this space.· Hiroki Kotabe, *The Prediction Market Primitive*
Integrating these AIs into the current prediction market framework can enable prediction markets at a microscopic scale, making them personally attractive and relevant.· Hiroki Kotabe, *The Prediction Market Primitive*
Where it matters
AI agents are the leverage layer for both liquidity (subsidized MMs in long-tail markets) and resolution (LLM judges for ambiguous outcomes). They're also the existential threat to today's retail trader edge · once agents become reliable, raw forecasting alpha disappears. For Dekant: an LLM-driven curve-drawing assistant would meaningfully lower the cognitive barrier to using continuous markets, where novice users currently can't translate a belief into a distribution shape.
Connections
- Liquidity provision · AI MMs as the subsidy mechanism for thin markets
- Oracle design · LLMs as judges in dispute resolution
- Forecasting accuracy · current AI agents underperform, but MCP improves
- Long-tail markets · where AI MMs can plausibly bootstrap liquidity
- Market manipulation · AI flow detection is a new surveillance challenge
Platforms linked to this concept
- Polymarket · affected-by · Cited as facing/exposed to AI agents
- ARENA (arenamarkets.fun) · implements · Mentioned in AI agents content as an implementing platform
- Kalshi · implements · Mentioned in AI agents content as an implementing platform
Related concepts
Sources
- Can LLMs Beat the Market? · OddChain · Mar 19 2026 ·
- Polymarket Doesn't Have a Money Problem. It Has a Plumbing Problem. · @allquantor · Mar 11 2026 ·
- Turning Probability into Assets: A Look Ahead at Prediction Market Agents · Jacob Zhao · Mar 5 2026 ·
- Prediction Markets are the Agentic Bazaar · Ben Fielding · Feb 16 2026 ·
- Building the Truth Machine · Andy Hall, Elliot Paschal · Feb 13 2026 ·
- Is AI Any Good at Predicting? · Mehmet Avci · Feb 2 2026 ·
- What to Do When Prediction Markets Fail · Andy Hall · Jan 24 2026 ·
- The Shape of Prediction Markets to Come · Will Owens · Jan 19 2026 ·
- The Prediction Market Primitive · Hiroki Kotabe · Apr 4 2024 ·