Jack Paul 
People have always argued about who will win an election or which celebrity couple will last. Putting real money behind those arguments is a different story. A functioning prediction market needs infrastructure for creating markets, matching trades, managing liquidity, settling winners, and dealing with disputes. For a long time, most of that infrastructure lived inside a few platforms. Users could trade on them, but building a new market platform from scratch was not realistic.
Rain Protocol is trying to change that with the second release of its SDK. The permissionless prediction market protocol now lets developers and AI agents launch independent prediction markets on networks like Arbitrum One. The v2 update includes migration guides for existing users moving from v1, so the change is aimed at both current builders and newcomers.
Less Infrastructure Work, More Product Work
The point of SDK v2 is to fold a lot of market infrastructure into the protocol itself. Market creation, trading, settlement, resolution, disputes, and appeals are all handled there now. That leaves developers free to focus on the product layer. They can decide what kinds of markets make sense, how users interact with them, and where a given market fits into the wider landscape. I think that shifts the focus from who can trade on prediction markets to who can build them.
The SDK is designed for both humans and AI agents. It has machine-readable documentation and built-in agent skills, so AI coding tools can understand the protocol and work with less hand-holding. That is timely. As AI agents take on more software work, they might eventually do more than help developers. They could help create markets too.
Trading Design Options
There is also a technical change on the trading side. Rain added an on-chain order book while keeping its existing automated market maker. Those two approaches suit different situations. An AMM can provide automated liquidity without needing a counterparty right away. An order book lets buyers and sellers place orders and have them matched on-chain. Builders can pick whichever structure fits their market’s expected liquidity and user behavior.
Other updates smooth out the user experience. A single session approval can cover multiple actions instead of requiring confirmation each time. Builders get real-time updates on trades and market activity. Users can convert collateral into Yes and No positions and back again without affecting the market price.
Wider Implications
Prediction markets may not stay inside a few big platforms forever. If the category keeps growing, the next phase may depend less on trading features and more on who is building markets. Rain’s latest SDK suggests the barrier to entry is starting to fall. Developers, and maybe the AI agents they train, are becoming a bigger part of how prediction markets take shape.