Terminal
Infiniti Feed

The Machines Are the Market Now: AI Agents Take Their Seats on the Book

Look at who actually makes money on the biggest prediction market and a pattern jumps out: fourteen of the twenty most profitable wallets on Polymarket are bots, according to FXVerify's reporting. One wallet turned roughly $300 into more than $400,000 in a single month by trading fifteen-minute Bitcoin, Ethereum and Solana contracts, exploiting the small lag between prices on the big crypto exchanges and prices on the prediction market. A 2025 study of a full year of Polymarket trading estimated arbitrage strategies extracted about $40 million from the platform. The fastest money in forecasting is no longer forecasting at all. It is plumbing.

That division of labor is becoming the market's structure. Machines excel at the mechanical edges: scanning order books for mispriced pairs, arbitraging the same event across venues, quoting both sides of thin markets and tightening spreads for everyone else. The infrastructure now assumes non-human customers. Polymarket's data and trading APIs are built for programmatic access, model-context-protocol servers stream live order books directly into AI agent frameworks, venues market themselves to autonomous agents, and purpose-built agents trade around the clock without sleep, fear or a favorite team.

But give the machines the actual forecasting job and the record humbles quickly. In the Prediction Arena experiment, six frontier AI models traded autonomously on Kalshi for 57 days and every one of them lost money, down between 16% and 31%. More research did not rescue them; the study found the volume of a model's research had little correlation with its returns. Meanwhile the profit distribution stays brutally concentrated: a Wall Street Journal analysis found about 67% of Polymarket's profit accrues to the top 0.1% of accounts, a leaderboard dominated by speed, size and systems rather than opinions.

The synthesis is the story. Prediction markets are becoming machine-readable probability infrastructure, a layer where prices are quoted in an API, consumed by software, and increasingly set by software too. That makes markets faster, tighter and harder to fool with stale prices. It does not make them oracles. The bots earn by correcting the price; the price still needs humans, with information and conviction, to be worth correcting toward.

For everyone else, the practical lesson is the same one Article One of this series delivered from the other side: know which game you are playing. Against a latency bot, a human clicking a price is the exit liquidity. In a slow, information-driven market, the machine is just another counterparty. The book now seats both species, and the edge belongs to whoever knows which seat they are in.

ai agentsbotsarbitragepolymarketkalshiInfiniti MarketsInfiniti Signals

Sources & attribution

Published
  1. Source 1
    FXVerify
    Publisher
    The publisher did not list a web address.
    Source URL
    The feed did not publish this story's article address.
  2. Source 2
    Prediction Arena study
    Publisher
    The publisher did not list a web address.
    Source URL
    The feed did not publish this story's article address.
  3. Source 3
    Wall Street Journal
    Publisher
    The publisher did not list a web address.
    Source URL
    The feed did not publish this story's article address.

Full attribution under the Digital Millennium Copyright Act (DMCA), 17 U.S.C. § 512: this article is an original report by Infiniti Intelligence that summarizes the reporting of the publishers named above, links to their original work, and reproduces no protected expression beyond brief factual reference and in-line attribution (“according to FXVerify”). All rights in the underlying reporting remain with the publishers listed; nothing here implies endorsement by or affiliation with them. Rights holders may request correction, update or removal through the contact address published at infiniti.w3ai.io.

Related