Imagine a future where your computer's tasks are handled seamlessly, almost autonomously, without you needing to lift a finger—this is exactly the promise that Simular’s AI technology is striving to deliver. But here’s where it gets controversial: Can AI truly take over complex tasks on your Mac or Windows PC in a way that’s both reliable and trustworthy? And this is the part most people miss—there are significant technical hurdles to overcome before this vision becomes a reality.
Simular, a promising startup, specializes in creating AI agents capable of controlling entire computers—something that sets it apart from many others that mainly focus on browser automation. Unlike typical AI systems that only handle web activities, Simular aims at managing the entire operating system. Their AI agent can mimic human actions like moving the mouse, clicking, copying, or pasting data—tasks that are often repetitive but necessary, such as populating spreadsheets with data.
Recently, Simular announced the official launch of its version 1.0 for Mac OS. The company is also collaborating with Microsoft to develop a similar AI agent for Windows, making it one of only five companies selected to join the Windows 365 for Agents program announced by Microsoft in mid-November. The other participants include Manus AI, Fellou, Genspark, and TinyFish. As for when the Windows version will be available, co-founder and CEO Ang Li indicated that they aim for it to be just as impactful and popular as the Mac version, though specifics remain somewhat vague.
What makes Simular particularly noteworthy is the impressive background of its founders. Ang Li, a scientist focused on lifelong learning, previously worked at Google’s DeepMind—a hub for cutting-edge AI research—where he met Jiachen Yang, their co-founder and reinforcement learning expert. They have a history of publishing influential papers, primarily aimed at advancing applications like Google’s self-driving car project, Waymo, rather than purely academic pursuits. This experience in practical, real-world AI solutions gives them a solid foundation for tackling the complex challenges ahead.
One major obstacle in advancing autonomous AI agents is the phenomenon called "hallucination"—where AI models generate inaccurate or misleading information. Large Language Models (LLMs), which underpin many AI systems, are known to hallucinate some of the time. For tasks involving thousands or even millions of individual steps, this becomes a critical issue, since an error at any point can invalidate the entire process. Statistically, the likelihood of hallucination errors increases as the number of steps involved grows.
A proposed solution—creating a deterministic version of the LLM—would ensure the AI’s responses or actions are the same every time, making behaviors predictable and reliable. However, this approach can limit the creative problem-solving capabilities of the AI, which is a core feature of agentic systems.
Simular is trying a hybrid approach. Its AI allows some flexibility in exploring different methods to complete a task, with human oversight to guide responses and correct mistakes during the process. Once a successful workflow is identified, the human user can lock it in—transforming that process into a deterministic script that can reliably be repeated. As Ang Li explains, "Our solution is to let agents keep exploring the successful trajectory, and once a reliable one is found, it becomes deterministic code."
This innovative process involves a technology that Li calls "neuro-symbolic computer use agents," which is different from the purely LLM-based systems used by many competitors. Instead of simply relying on language models to generate responses, their system has the AI write code that becomes deterministic—ensuring consistent outcomes for repeated tasks. With this, the user gains full transparency and control, as they can inspect, audit, and understand the code behind their automation workflows.
While still in early stages, early beta users of Simular include a car dealership automating VIN searches and homeowners’ associations extracting contract data from PDFs. Their open-source project, although currently available only for Mac OS, has already enabled automations across various domains such as content creation, marketing, and sales.
Having previously raised a $5 million seed round, bringing their total funding to roughly $27 million, Simular continues to attract notable investors including Basis Set Ventures, Flying Fish Partners, Samsung NEXT, Xoogler Ventures, and angel investor Lenny Rachitsky.
So, will innovative technological solutions like Simular’s become the standard for workplace automation? Are we heading toward a future where AI can reliably perform complex routines without constant human oversight? Share your thoughts—do you believe this is the breakthrough that will democratize AI-powered PC control, or are there hurdles that still need addressing before this vision can unfold fully? Let’s start the conversation.
