AI-run prop firms tighten markets on Kalshi and Polymarket

Prop trading firms are deploying AI agents and wiring professional liquidity into event contracts on Kalshi and Polymarket. Market makers, quantitative shops and funded traders are quoting both sides of contracts, comparing prices across venues and repricing continuously.

A July 21 poll of 104 economists found all expected the Federal Reserve to hold its policy rate at 3.50%–3.75% at the July 28–29 meeting. Kalshi’s July contract priced that outcome at about 87% with roughly $29.7 million displayed in active volume.

Combined monthly volume across Kalshi and Polymarket peaked at $13.7 billion in June and exceeded $11 billion in July. Kalshi reported annualized volume near $178 billion and said institutional flow rose about 800% over six months.

Firms building access and infrastructure around those markets include Clear Street, Marex and Jump Trading. AQR, Susquehanna and OKX have advertised specialist roles tied to prediction markets. Corporate treasuries are testing event contracts to hedge tariff and regulatory exposure.

Proprietary trading shops, market makers and algorithmic firms are using automated strategies and AI agents to evaluate signals and convert forecasts into trades. The format of binary event contracts lets allocators observe probability calibration and outcomes directly.

One benchmark estimates about 350 resolved binary predictions are needed to detect a two percentage point edge with reasonable confidence, and roughly four times that to confirm a one‑point edge. Louis Régis, founder of on‑chain prop firm Propr, described himself as confident about the direction but not the magnitude.

Propr plans to extend its evaluation model to Polymarket, allowing traders and AI agents to qualify for accounts up to $100,000 and hold up to $300,000 across accounts, with an 80% profit share after qualification. The firm currently routes roughly 5% of its signals to live venues and simulates the rest, with payouts settling in USDC.

A test that let six frontier models trade autonomously with $10,000 each from Jan. 12 to Mar. 9 returned losses between 16% and 30.8% on Kalshi and an average loss near 1.1% on Polymarket. A working paper on execution notes that forecasting accuracy requires a betting strategy and venue liquidity to translate into profit.

Market participants identify the next tests as the Fed decision on July 28–29, the Bureau of Economic Analysis’ advance GDP estimate on July 30 and the July employment report on Aug. 7. Those releases will prompt repricing across event contracts and further activity by funded traders and AI agents.