Are you seeing this movement of agents in prediction markets? Well, what has been fermenting since last year is finally taking shape in 2026. It’s no longer just theory.

The numbers speak for themselves: predictive markets grew insanely from 9 billion in 2024 to over 40 billion in 2025. Over 400% annual growth. Polymarket and Kalshi are practically dominating the space now, with Kalshi already surpassing Polymarket in volume in February. The difference is that Kalshi entered through the traditional regulatory path while Polymarket continues on the decentralized route.

But the interesting point isn’t just the market growth. It’s how agents are starting to operate in these markets automatically. The real advantage of these agents isn’t “predicting more accurately than humans.” It’s processing information faster and executing with discipline that no trader can maintain 24/7.

Think about it: prediction markets are basically machines that aggregate dispersed information into price signals. The more people participating, the more efficient it becomes. That’s where the agent comes in: it scans news data, social media, blockchain, identifies pricing deviations, and executes. All of this in seconds.

The architecture of these agents follows a well-defined pattern. Data collection layer (gathers data), analysis layer (identifies opportunities), strategy layer (sets positions), and execution layer (performs transactions). Simple in theory, complex in practice.

Now, about position management: many people think Kelly’s formula is the magic solution. But in practice, professional traders use variations of it. The original Kelly criterion requires precise probability estimates that are hard to maintain consistently. So what works best is a stepwise approach: you classify opportunities into confidence levels and assign each level a fixed position size. No false precision, no mathematical illusions.

Some people use pure Kelly criterion, some use flat betting, some use inverted risk limits. The important thing is that the strategy is codifiable and testable. Most agents I’ve seen out there are still far from that.

In terms of strategy, deterministic arbitrage is where agents shine. Arbitrage (when the outcome is almost decided but the market hasn’t priced it yet), Dutch Book arbitrage (when the sum of probabilities doesn’t add up), platform-to-platform arbitrage. These are clear rules, low-risk, fully automatable.

Directional speculation is more complicated. It requires judgment, context interpretation, and then the agent becomes limited. It can be used as a complement but not as the main strategy.

The projects I see in the market now? Olas Predict is probably the most advanced in terms of product. Polyseer is doing structured analysis with multiple agents. Verso is a Bloomberg-style terminal for institutional traders. Matchr does intelligent routing between platforms. Each trying to solve a different piece of the puzzle.

But here’s the truth: there’s still no standardized, mature product that integrates everything well. Strategy generation, execution efficiency, risk control, closed commercial cycle. It’s all still fragmented.

The most sensible business model is a three-layer one: infrastructure (sells data and APIs to B2B), strategies (monetize through commissions or revenue sharing), and managed agents (management fee + performance). That way, you’re not solely dependent on beating the market. If alpha contracts, you still have revenue from infrastructure and tools.

Honestly, I think 2026 will be the year these prediction market agents move out of the experimental phase and start operating for real. But they still lack maturity. Lack standardization. Lack a track record of consistent performance. Whoever manages to solve that will have a very interesting business in hand.