Reflection AI founded

Misha Laskin and Ioannis Antonoglou founded Reflection AI, a startup aimed initially at building autonomous coding agents. Both had come from Google DeepMind: Antonoglou had spent more than a decade there and was one of the core architects of AlphaGo, the system that beat a world-champion Go player in 2016; Laskin had worked on reward modelling for the Gemini project after a doctorate under Pieter Abbeel at Berkeley.

The founders’ pitch drew on their reinforcement-learning background rather than pure language-model scaling: agents that could plan, act and learn from the outcome of their actions in a codebase, rather than simply predicting the next token. This placed the company in a wave of 2024 startups betting that coding was the task where autonomous, tool-using agents would first become commercially useful.

Reflection AI’s ambitions widened considerably after founding. By late 2025 the company had repositioned as an attempt to build an open-weight frontier lab to rival Chinese developers such as DeepSeek, raising a $2 billion round at an $8 billion valuation — a scale of funding unusual even by the standards of the period, and a marker of how far investor appetite for foundation-model startups had grown in the year and a half since a small team of ex-DeepMind researchers set out to automate software engineering.