Simular Secures $21.5M for PC-Controlling AI Agents

Simular, a pioneering startup developing AI agents for both Mac OS and Windows, has successfully secured $21.5 million in a Series A funding round. This latest investment, led by Felicis, brings the company's total capital raised to approximately $27 million. Simular distinguishes itself by focusing on direct PC control rather than just browser automation, and critically, by introducing an innovative method to tackle the pervasive problem of AI hallucinations in complex agentic tasks.

Funding Fuels Advanced Desktop AI Automation

The Series A round saw participation from existing seed investors, including NVentures (NVIDIA’s venture arm), South Park Commons, Basis Set Ventures, Flying Fish Partners, Samsung NEXT, Xoogler Ventures, and prominent angel investor Lenny Rachitsky. This significant funding underscores investor confidence in Simular's vision for advanced desktop AI automation.

Beyond the Browser: Direct PC Control

Unlike many other agentic AI startups that primarily focus on browser-based interactions, Simular aims to empower AI agents to control the entire personal computer. Co-founder and CEO Ang Li explained to TechCrunch:

"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."

He cited copying and pasting data into a spreadsheet as a simple example of the agent's capabilities. Agentic AI refers to systems designed to autonomously complete complex tasks with minimal human intervention.

Platform Rollout and Microsoft Partnership

Simular recently announced the release of its 1.0 version for Mac OS. Concurrently, the startup is actively collaborating with Microsoft to develop an agent for Windows. Simular is one of only five agentic companies selected for the exclusive Windows 365 for Agents program, which Microsoft unveiled in mid-November. While CEO Li remained vague about the exact timeline for the Windows version, he indicated it is poised to be as popular, if not more so, than its Mac counterpart. The other companies in this prestigious program include Manus AI, Fellou, Genspark, and TinyFish.

Solving the AI Hallucination Challenge

A major hurdle for the widespread adoption of agentic AI is the issue of large language model (LLM) hallucinations, where AI generates incorrect or nonsensical information. For agentic tasks requiring thousands or even millions of discrete steps, a single hallucination can invalidate an entire workflow, with the probability of error increasing with task complexity.

Traditional approaches to combat this involve making LLMs "deterministic" by scripting responses, which can limit their creative problem-solving abilities. Simular, however, has devised a novel solution. Its agent iterates freely on a task, with a human user providing real-time course corrections until success is achieved. Once a successful trajectory is found, the human user "locks in" that workflow, transforming it into deterministic, repeatable code.

Li elaborated:

"Our solution is, let agents keep exploring the successful trajectory. Once you found a successful trajectory, that becomes deterministic code."

Unique Technology and User Trust

Simular's approach is not merely an LLM wrapper. Li states:

"We have a new technology which is not used by any other agent company. We call it ‘neuro symbolic computer use agents.’ It’s not fully LLM based."

He further explained:

"Our approach to solve hallucinations is to let the LLM write code which becomes deterministic. So if you have a workflow that works, the next time we run the same workflow, it’ll be successful as well."

A significant benefit of this method is that the deterministic code, which performs repeatable tasks, resides with the end user, not solely within the LLM. This empowers users to inspect and audit the code, fostering trust.

"Once they have the code, they can trust it, because they can inspect it, they can audit it, they can see what’s going on," Li affirmed.

Early Impact and Future Potential

While acknowledging the early stage of their work, Li shared promising examples from beta customers. These include a car dealership automating VIN number searches and Homeowners Associations (HOAs) efficiently extracting contract information from PDFs. Furthermore, Simular's open-source project, currently available for Mac OS, has already facilitated automations across various domains, from content creation to sales and marketing.

Founding Team's Pedigree

The credibility of Simular's founders further strengthens its position. Ang Li, a continuous learning scientist, previously worked at Google's DeepMind. There, he met co-founder Jiachen Yang, a reinforcement learning specialist. Their collective experience at DeepMind involved not just academic research but also practical application, contributing to the improvement of Google products like Waymo, providing a strong foundation for their agentic AI venture.