There is something interesting happening in the prediction markets that I think is worth following.

In 2024, the total trading volume was around US$ 9 billion. It jumped to over US$ 40 billion in 2025 — growth of over 400%. It’s no exaggeration to say this has become an irreversible trend. Polymarket and Kalshi dominate the space, but the dynamics between them are interesting: Kalshi emerged victorious in legal issues regarding electoral contracts and expanded quickly into sports markets; Polymarket built a more decentralized and global architecture, "off-chain matching, on-chain settlement."

But what really caught my attention is the potential of AI agents in this space. And here’s the crucial point: it’s not about AI "predicting better" than humans. It’s about transforming dispersed information into opportunities for faster and more disciplined execution. Prediction markets, by nature, aggregate information through real transactions — it’s a kind of positive externality of the system. When you have this structure, agents can exploit inefficiencies much more efficiently than manual traders.

The architecture makes sense: information layer (news, on-chain data), analysis layer (identify price deviations), strategy layer (calculate optimal position), execution layer (multiple markets, slippage optimization). The challenge is that not all strategies work well with automation. Liquidation arbitrage — when the outcome is already determined but the market hasn’t priced it yet — is practically designed for agents. Platform arbitrage as well. But directional speculation? That still requires human judgment.

In terms of position management, the Kelly formula is the classic theory, but in practice, professional traders use something simpler: they divide capital into fixed units and vary the number of units based on confidence in the signal. This reduces complexity and leaves less room for model errors.

In the current landscape, we have official frameworks (Polymarket launched its own agent framework), analysis tools like Polyseer and Oddpool, and some autonomous agents like Olas Predict and UnifAI Network. But honestly, we haven’t seen a mature product that integrates everything — strategy generation, efficient execution, systematic risk control, and a closed business model. Olas Predict is probably the most advanced, but still limited by Omen’s liquidity.

In terms of monetization, I see three paths: infrastructure (data and B2B tools), strategy subscription (SaaS signals), and managed vaults (more complex regulation). The most viable path now is the middle one — signal tools without custody of funds. Less regulatory friction, more predictable revenue.

What makes all this interesting is that prediction markets are evolving from a niche to something with real externalities in the financial system — CME and Bloomberg are integrating probabilities of events as market data. When you have infrastructure, liquidity, and regulatory clarity (that the US is starting to offer), AI agents gain legitimate space to operate.

We’re still early, but the space is moving quickly. It’s worth following not only the agents themselves but also analysis tools and the strategies that work best. If you’re in Gate, with access to data from multiple chains — it could be a good starting point to explore how agents could function in decentralized prediction markets.