Simular, a startup constructing AI brokers for Mac OS and Home windows, has raised a $21.5 million Collection A led by Felicis, with present seed buyers NVentures (NVIDIA’s enterprise arm), South Park Commons, and others becoming a member of in.
Simular is an attention-grabbing agentic startup as a result of in contrast to others, it isn’t attempting to manage the browser however the PC itself. (Agentic AI refers to programs that may autonomously full advanced duties with minimal human intervention.) “We will actually transfer the mouse on the display and do the clicking. So it’s extra able to doing, repeating no matter human actions within the digital world,” co-founder CEO Ang Li advised TechCrunch, providing the instance of copying and pasting knowledge right into a spreadsheet.
On Monday it introduced the discharge of its 1.0 model for Mac OS. Nevertheless it’s additionally working with Microsoft to develop an agent for Home windows. The startup is certainly one of 5 agentic corporations accepted into the Home windows 365 for Brokers program Microsoft introduced in mid November. (The others are Manus AI, Fellou, Genspark, and TinyFish.) As for the timeline for the Home windows model, Li was obscure besides to say it guarantees to be as or extra standard than the Mac model.
One more reason to look at Simular is the bona fides of the founders: Li is a steady studying scientist who beforehand labored at Google’s DeepMind, the place he met his cofounder, reinforcement studying specialist Jiachen Yang. Whereas their staff printed their fair proportion of papers, the work wasn’t strictly tutorial, Li mentioned. It was supposed to enhance Google merchandise, together with Waymo.
That AI product background is useful as a result of, earlier than the agentic way forward for Silicon Valley’s desires can materialize, there are a number of technical issues to unravel. One of many largest is that LLMs hallucinate some share of the time.
Agentic duties can require finishing hundreds to thousands and thousands of discrete steps. Not solely can a hallucination at any single step invalidate the entire agent’s work, however hallucinations develop into statistically extra possible because the variety of steps grows.
One method to clear up that is to make the “non deterministic” LLM “deterministic,” which means as an alternative of permitting the LLM to be endlessly inventive, its responses or actions are scripted the identical every time. However that dangers limiting the entire inventive downside fixing side of an agent.
Techcrunch occasion
San Francisco
|
October 13-15, 2026
Simular is marrying the 2. Its agent will iterate freely on the duty, with the human person within the center course correcting, till the agent achieves success. Then the human locks in that job’s workflow, which makes it deterministic, repeatable.
“Our answer is, let brokers maintain exploring the profitable trajectory. When you discovered a profitable trajectory, that turns into deterministic code,” Li explains.
The rationale the startup can do it’s because its work – which Li admits remains to be early – is not only an LLM wrapper that sends and retrieves knowledge to a mannequin.
“We’ve a brand new know-how which isn’t utilized by another agent firm. We name it ‘neuro symbolic pc use brokers.’ It’s not totally LLM based mostly,” he mentioned. “Our strategy to unravel hallucinations is to let the LLM write code which turns into deterministic. So if in case you have a workflow that works, the following time we run the identical workflow, it’ll achieve success as effectively.”
One other profit is that this deterministic code that performs a repeatable job is within the arms of the tip person, not the LLM. “As soon as they’ve the code, they will belief it, as a result of they will examine it, they will audit it, they will see what’s occurring,” Li says.
Time will inform if this methodology is the magic that can convey brokers into the arms of each employee. Li says his early beta clients embody a automobile dealership automating VIN quantity searches, and HOAs extracting contract info from PDFs. And the corporate’s open supply challenge (obtainable just for Mac OS for the time being) has led to automations starting from content material creation to gross sales and advertising.
Simular beforehand raised a $5 million seed, bringing its whole raised to about $27 million. Different buyers within the firm embody Foundation Set Ventures, Flying Fish Companions, Samsung NEXT, Xoogler Ventures, and podcaster and angel investor Lenny Rachitsky, the corporate says.
Keep forward of the curve with NextBusiness 24. Discover extra tales, subscribe to our publication, and be a part of our rising neighborhood at nextbusiness24.com
