The AI agent wars just got personal - literally. Simular closed a $21.5 million Series A led by Felicis to fund its desktop automation agent that doesn't just browse the web but actually controls your entire Mac or Windows PC. The startup claims it's cracked the code on AI hallucinations with a hybrid approach that lets agents learn tasks, then locks successful workflows into reliable, repeatable code.
Simular just positioned itself at the center of the AI agent revolution with a $21.5 million Series A that could reshape how we think about desktop automation. The startup, led by ex-Google DeepMind scientists, isn't content with browser-based agents - they want to control your entire computer.
The funding round, led by Felicis, brings Simular's total raised to $27 million and validates a radically different approach to AI agents. While competitors focus on web automation, Simular's agent literally moves your mouse cursor and clicks buttons across any application on your desktop. "We can literally move the mouse on the screen and do the click. So it's more capable of doing, repeating whatever human activities in the digital world," CEO Ang Li told TechCrunch.
The timing couldn't be better. Microsoft just selected Simular as one of five companies for its Windows 365 for Agents program, alongside Manus AI, Fellou, Genspark, and TinyFish. The partnership signals Microsoft's serious push into agentic computing, with Simular's Windows version expected to match or exceed the popularity of its Mac release.
But Simular's real breakthrough isn't just desktop control - it's solving the hallucination problem that's plagued AI agents since day one. Current large language models hallucinate unpredictably, and when agents need to execute thousands of steps, even small errors cascade into complete failures. Traditional solutions force deterministic behavior that kills creativity, or accept unreliable performance.
Li and co-founder Jiachen Yang, both reinforcement learning specialists from DeepMind's product teams (including Waymo development), engineered a hybrid solution they call "neuro symbolic computer use agents." The system lets AI agents explore and iterate freely on tasks with human feedback, but once a successful workflow emerges, it converts that process into deterministic code.












