Why a Y Combinator startup tackling AI brokers for Home windows gave up and pivoted

0
13
Why a Y Combinator startup tackling AI brokers for Home windows gave up and pivoted



A startup referred to as Pig.dev that participated in Y Combinator’s Winter 2025 batch was engaged on a possible revolutionary thought: AI agentic tech to regulate a Microsoft Home windows desktop.

However in Could, the founder introduced he was abandoning the tech and pivoting his firm to one thing solely completely different: Muscle Mem, a cache system for AI brokers that permits them to dump repeatable duties. 

An early-stage YC firm pivoting is nothing out of the atypical, after all. What’s fascinating — and what sparked a dynamic dialog on Thursday’s Y Combinator podcast — is that Pig was engaged on pc use, one of many massive areas that must be solved for brokers to be actually helpful within the workforce. One other firm – and one other YC alum – that’s tackling that for the browser is named Browser Use.

Browser Use surged to reputation when Chinese language agentic device Manus, which relied on it, went viral. Browser Use basically scans the buttons and parts of a web site to show them right into a extra digestible, “text-like” format for brokers, serving to the AI perceive the way to navigate and use the web site.

In the course of the Y Combinator podcast, launched Thursday, companion Tom Blomfield likened Pig because the Browser Use for Home windows desktops. The podcast featured Amjad Massad, the founder and CEO of standard vibe-coding startup Replit.

Massad, Blomfield, and YC companion David Lieb had been discussing how long-term pc use of hours, slightly than minutes, was nonetheless a stumbling block for brokers. Because the context window for reasoning grows, an agent’s accuracy wavers whereas LLM prices improve.

“The recommendation I’d give founders right now is taking both Browser Use or Home windows automation with Pig and attempting to use that into enterprise, right into a vertical business,” Blomfield instructed. 

Techcrunch occasion

San Francisco
|
October 27-29, 2025

Massad agreed, “The second that know-how works, these two firms are going to do actually, rather well,” he mentioned.

However alas, Pig’s founder Erik Dunteman has already given up on the thought. In his put up in Could he defined that he at first needed to run a cloud API product (a standard approach of delivering AI tech). However his prospects didn’t need that. So he tried promoting it as a dev device. And so they didn’t need that both. 

“What customers within the legacy app automation house really need is handy me cash, and obtain an automation,” he mentioned. Basically, they needed to rent a marketing consultant to make their desired Home windows robotic course of automations work for them. 

However Dunteman didn’t wish to do one-off tasks. He needed to construct improvement instruments. So he deserted Pig and began engaged on an AI caching device. Dunteman declined additional remark about his resolution to ditch Home windows automation, though the Pig.dev web site and Github paperwork stay accessible. 

Nonetheless, Dunteman did inform us his new device was impressed by the pc use drawback. It’s chipping away at it from one other angle. The thought is to permit the agent to dump repeated duties to the Muscle Mem service, so the agent can concentrate on reasoning for brand new issues and edge instances.

“What we’re engaged on now’s straight impressed by and relevant to pc use, simply on the developer tooling layer. I stay very optimistic for pc use as ‘the final mile’,” he instructed TechCrunch.

That’s to not say that nobody is engaged on Home windows automation.

In all probability the corporate farthest alongside on that’s Microsoft itself. As an example, in April Microsoft introduced it added pc use tech to Copilot Studio for graphical person interfaces like Home windows. That tech was launched as a analysis preview.  Plus, earlier this month, Microsoft introduced an agentic device in Home windows 11 that helps finish customers handle settings.