These prediction markets that we see growing exponentially in 2025 and 2026 are not just a passing trend anymore. I have been closely monitoring them, and the transformation is real: we went from approximately $9 billion in annual volume in 2024 to over $40 billion in 2025. That’s more than 400% growth.

But what really caught my attention is that these markets have stopped being just betting tools. They evolved into something much more interesting: a global layer of truth. And here’s the positive asymmetry that few realize: when you aggregate dispersed information through real transactions with money at stake, the resulting price carries much more collective wisdom than any other isolated source.

I now see the emergence of AI agents operating in these markets, and honestly, their value isn’t in “predicting better than humans.” It’s in processing information faster, with more discipline and without emotion. The positive asymmetry these agents exploit is precisely the temporal inefficiency: when news breaks, there’s a gap between the moment it should be priced in and when the market actually incorporates it. Agents fill that gap.

Polymarket and Kalshi are practically dominating the space now. In February 2026, Kalshi already had $25.9 billion in weekly volume, surpassing Polymarket which was at $18.3 billion. The strategic difference is clear: Polymarket built a decentralized global market, while Kalshi integrated into the traditional American financial system.

What interests me most is how the agents are being structured. I’ve seen several attempts in the market, but most are still far from a mature product. Olas Predict was one of the first to take this seriously with Omenstrat, and recently they launched Polystrat for Polymarket. UnifAI Network has a very specific approach: buying contracts near settlement with implied probabilities above 95%, seeking spreads of 3-5%.

But here’s the point that makes me think: the true positive asymmetry for these agents doesn’t just come from speed. It comes from the ability to execute deterministic strategies in environments with clear rules. Settlement arbitrage, Dutch Book arbitrage, platform-to-platform trading — these are the ones that really work well in automation.

The challenge I see is that as more agents enter the space, this positive asymmetry will erode. That’s why the most sustainable business models don’t bet everything on alpha. They create layers: data and execution infrastructure (stable B2B revenue), third-party strategy ecosystems (monetization via fees or equity), and vaults with autonomous agents (management and performance fees).

I’ve been following tools like Polyseer, Oddpool, and Verso. These platforms are building the analysis and signal foundation that agents need. But they still lack that leap to a truly closed product, with integrated systematic risk management, robust backtesting, and reliable execution.

The market is still in a very early stage. But the direction is clear: prediction market agents will be an important product form in the coming years, as long as they can maintain that positive asymmetry through disciplined execution, low operational costs, and the ability to capture intermarket opportunities. Those who manage to combine solid infrastructure with real risk control will be well positioned.