I just read a fairly in-depth analysis about the Prediction Market Agents trend in 2026, and there are a few noteworthy points here.

The main thing is that prediction markets have become a real phenomenon. In 2024, the trading volume was only about $9 billion, but by 2025 it surged to over $40 billion—more than a 400% growth in one year. This figure is no coincidence: Polymarket and Kalshi dominate nearly the entire market, with Kalshi recently surpassing Polymarket in trading volume thanks to recent legal wins in the U.S.

But what’s more interesting is how Prediction Market Agents are forming. Instead of just being "better prediction AIs," they are essentially probability portfolio management systems—transforming news and on-chain data into trading opportunities that they can execute quickly, disciplined, and at lower costs than humans.

The ideal structure of such an agent has four layers: the information layer (collects data from multiple sources), the analysis layer (detects mispricings), the strategy layer (converts opportunities into positions), and the execution layer (places orders, optimizes slippage). It sounds simple but is much more complex in practice.

Regarding strategies, I see that not all types of trading are suitable for automation. Arbitrage detection—such as payment discrepancies or platform-to-platform differences—is the most suitable because the rules are clear. Meanwhile, speculative trading based on judgment and information still requires human intervention.

An important point emphasized in the report is position management. Instead of mechanically applying the Kelly Criterion (because it is very sensitive to errors in probability estimates), real-world agents should use simpler methods: dividing capital into fixed units, or using discrete confidence levels with fixed position limits. This reduces risk and increases fault tolerance.

On the business model, I see three main paths: infrastructure layer (providing data and execution tools, charging B2B), strategy layer (allowing community contributions and revenue sharing), and Vault (asset management agents with management fees + performance fees). Currently, the "strategy subscription + signals" positioning is the most feasible because it doesn’t face overly heavy legal barriers.

The current state of the ecosystem is also interesting: Polymarket has officially released the Polymarket Agents framework, Gnosis has PMAT, while Kalshi is still at the SDK level. Olas Predict is the most advanced product today, recently launching Polystrat to expand into Polymarket. But honestly, no product has yet fully integrated strategy generation, risk control, and sustainable business cycles.

Overall, Prediction Market Agents are still in the exploratory phase. Prediction markets are not gambling—they are a "global truth" layer where events are publicly priced through collective intelligence. These agents will be tools to systematically exploit this market efficiently, as long as they focus on what they do best: fast data processing, disciplined execution, and tight risk management.

Is anyone else following these projects? I’m curious to see what developments will happen in the next six months.