AnyAlpha turns a trading thesis into an agent that acts

A trading idea can start as a sentence. AnyAlpha wants to turn that sentence into software that finds relevant markets, tests a strategy against historical data and executes trades when its conditions are met.

The startup has launched an agentic trading platform, initially focused on prediction markets, with an iOS app linked from its announcement. Users discuss an idea with the product, refine it and let the system build an agent around it. The company says the agent can then monitor markets and trade across multiple exchanges around the clock.

The interesting step is the translation between an informal belief and an executable strategy. A person may think an event is more likely than its market price suggests, but putting that view into practice requires deciding which markets matter, what information to monitor and when to enter or exit. AnyAlpha says it combines research, historical testing and execution in one workflow.

Its launch illustrates the process with a user's thesis that unusually hyped NFL teams are overpriced. The described agent follows social attention, compares prediction-market prices with estimated win probabilities, defines trading conditions and tests the approach across earlier seasons. The company names Kalshi and Polymarket in that example.

The example explains the product's intended behavior; it does not establish that the strategy earns a profit. Estimated probabilities can be wrong, and a backtest describes results under historical assumptions rather than proving how an agent will perform with live money.

Watch AnyAlpha's launch video.

The company positions prediction markets as its starting point because information relevant to an outcome can be scattered across news, data services and social platforms. Its broader ambition is to support other asset classes. That expansion is a plan, while prediction-market research and trading are the focus of the current announcement.

Founders Shirley (Shiyi) Peng and Jerry Miao met at a poker tournament at Microsoft, according to the launch. Peng studied computer science at the University of Waterloo and worked on trading infrastructure at Coinbase and a trading terminal at LSEG. Miao studied computer science at UC Berkeley and has worked on his own quantitative trading systems.

The founders say they previously bootstrapped multiple apps to $500,000 in annual recurring revenue. That figure refers to their earlier applications, not reported revenue for AnyAlpha. The company is part of Y Combinator's Fall 2026 batch.

AnyAlpha also describes the possibility of creating, running and combining multiple agents. The launch does not provide enough detail to assess every execution safeguard or the availability of each exchange integration for every user, so those remain questions for the product rather than established features to assume.

The practical shift is from a tool that returns an answer to one that can keep acting on a thesis after its user leaves the screen. For that workflow, understanding the agent's conditions and reviewing what its tests actually demonstrate matter as much as generating the initial strategy. A convincing explanation of a trade is still different from evidence that it should be executed.

‍