Simular, a startup building AI agents for Mac OS and Windows, has raised a $21.5 million Series A led by Felicis, with existing seed investors NVentures (NVIDIA’s venture arm), South Park Commons, and others joining in.
Simular is an interesting agentic startup because unlike others, it isn’t trying to control the browser but the PC itself. (Agentic AI refers to systems that can autonomously complete complex tasks with minimal human intervention.) “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,” co-founder CEO Ang Li told TechCrunch, offering the example of copying and pasting data into a spreadsheet.
On Monday it announced the release of its 1.0 version for Mac OS. But it’s also working with Microsoft to develop an agent for Windows. The startup is one of five agentic companies accepted into the Windows 365 for Agents program Microsoft announced in mid November. (The others are Manus AI, Fellou, Genspark, and TinyFish.) As for the timeline for the Windows version, Li was vague except to say it promises to be as or more popular than the Mac version.
Another reason to watch Simular is the bona fides of the founders: Li is a continuous learning scientist who previously worked at Google’s DeepMind, where he met his cofounder, reinforcement learning specialist Jiachen Yang. While their team published their fair share of papers, the work wasn’t strictly academic, Li said. It was intended to improve Google products, including Waymo.
That AI product background is helpful because, before the agentic future of Silicon Valley’s dreams can materialize, there are a host of technical problems to solve. One of the biggest is that LLMs hallucinate some percentage of the time.
Agentic tasks can require completing thousands to millions of discrete steps. Not only can a hallucination at any single step invalidate all of the agent’s work, but hallucinations become statistically more likely as the number of steps grows.
One way to solve this is to make the “non deterministic” LLM “deterministic,” meaning instead of allowing the LLM to be endlessly creative, its responses or actions are scripted the same each time. But that risks limiting the whole creative problem solving aspect of an agent.
Techcrunch event Join the Disrupt 2026 Waitlist Add yourself to the Disrupt 2026 waitlist to be first in line when Early Bird tickets drop. Past Disrupts have brought Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla to the stages — part of 250+ industry leaders driving 200+ sessions built to fuel your growth and sharpen your edge. Plus, meet the hundreds of startups innovating across every sector. Join the Disrupt 2026 Waitlist Add yourself to the Disrupt 2026 waitlist to be first in line when Early Bird tickets drop. Past Disrupts have brought Google Cloud, Netflix, Microsoft, Box, Phia, a16z, ElevenLabs, Wayve, Hugging Face, Elad Gil, and Vinod Khosla to the stages — part of 250+ industry leaders driving 200+ sessions built to fuel your growth and sharpen your edge. Plus, meet the hundreds of startups innovating across every sector. San Francisco | WAITLIST NOW
Simular is marrying the two. Its agent will iterate freely on the task, with the human user in the middle course correcting, until the agent achieves success. Then the human locks in that task’s workflow, which makes it deterministic, repeatable.
“Our solution is, let agents keep exploring the successful trajectory. Once you found a successful trajectory, that becomes deterministic code,” Li explains.
... continue reading