Simular’s AI agent wants to run your Mac, Windows PC for you

Simular, an innovative startup developing AI agents for both Mac OS and Windows, has successfully closed a $21.5 million Series A funding round. The investment was spearheaded by Felicis, with participation from existing seed backers including NVentures, the venture arm of Nvidia, and South Park Commons. What sets Simular apart in the crowded AI landscape is its ambition to manage entire PCs rather than just web browsers. Co-founder and CEO Ang Li explained to TechCrunch, "We can literally move the mouse on the screen and perform clicks, enabling the AI to replicate a wide array of human activities in the digital environment." An example he provided involved automating tasks like copying and pasting data into spreadsheets. On the heels of announcing its 1.0 version for Mac OS, the company is also collaborating with Microsoft to create an agent tailored for Windows platforms. Simular is one of five companies selected for the Windows 365 for Agents program, which Microsoft unveiled in mid-November. Li hinted that the Windows version could match or exceed the popularity of its Mac counterpart, though specific timelines remain unclear. The credibility of Simular's leadership enhances its potential. Li, a continuous learning scientist, previously contributed to Google’s DeepMind, where he crossed paths with co-founder Jiachen Yang, an expert in reinforcement learning. Their collective experience has not only led to academic publications but has also focused on enhancing Google's products, including the autonomous driving technology, Waymo. As the startup navigates the challenges of agentic AI, it faces significant technical hurdles. One major issue is the tendency of large language models (LLMs) to produce inaccuracies, known as hallucinations. Given that agentic tasks can involve thousands or even millions of individual steps, a hallucination at any point can jeopardize the entire operation. To address this, Simular aims to transform the 'non-deterministic' nature of LLMs into a 'deterministic' process, where responses are predictable and consistent. The innovative approach Simular is pursuing allows agents to explore various methods to accomplish a task, with human users providing guidance and corrections along the way. Once a successful workflow is established, it becomes repeatable and reliable. "Our solution is to let agents keep exploring successful paths. Once a successful trajectory is found, that becomes deterministic code," Li elaborated. Crucially, this deterministic code is placed in the hands of the end-users, allowing for transparency and trust, as they can inspect and audit the processes. Early beta testers include a car dealership automating VIN number searches and homeowners' associations parsing contract details from PDFs. Currently, Simular's open-source project, available for Mac OS, has led to a spectrum of automations, spanning content creation to sales and marketing. With a total funding of approximately $27 million, following a previous $5 million seed round, Simular is backed by notable investors, including Basis Set Ventures, Flying Fish Partners, Samsung NEXT, Xoogler Ventures, and angel investor Lenny Rachitsky. As the company continues its journey, the potential for its technology to empower workers in various sectors remains to be seen.