In a recent turn of events, Pig.dev, a startup that participated in Y Combinator’s Winter 2025 batch, has made a significant pivot in its business strategy. Originally, the company was developing a potentially revolutionary AI agentic technology designed to control Microsoft Windows desktops. However, in May, founder Erik Dunteman announced that he would be abandoning this endeavor to focus on a new project: Muscle Mem, an innovative caching system for AI agents that facilitates the offloading of repeatable tasks.
It’s not uncommon for early-stage startups, especially those in Y Combinator, to pivot away from their initial ideas. What sets Pig.dev's transition apart is its initial focus on computer use, a critical area that remains largely unresolved for AI agents to be genuinely effective in the workplace. This topic sparked a lively discussion during the Y Combinator podcast released on Thursday, highlighting the challenges and opportunities in the realm of AI.
Another Y Combinator alum, Browser Use, is addressing similar challenges within the browser environment. This company gained traction when the Chinese agentic tool Manus, which utilized Browser Use technology, went viral. Essentially, Browser Use scans the buttons and elements of websites, transforming them into a more digestible, “text-like” format. This adaptation allows AI to navigate and interact with websites more effectively, enhancing the overall user experience.
During the podcast, Y Combinator partner Tom Blomfield drew a parallel between Pig.dev and Browser Use, suggesting that Pig could become the equivalent for Windows desktops. The discussion also featured Amjad Masad, founder and CEO of the popular coding platform Replit, who echoed the sentiment that long-term computer use remains a significant hurdle for AI agents. As the context window for reasoning expands, an agent’s accuracy can diminish, leading to increased costs associated with large language models (LLMs).
Blomfield offered advice for aspiring founders, encouraging them to consider either Browser Use or Windows automation technologies—such as Pig’s original concept—and apply them to enterprise solutions or specific verticals. Masad concurred, stating that once the technology matures, both companies could achieve substantial success.
Despite the promising discussions, Erik Dunteman has chosen to abandon his original vision for Pig.dev. In a post from May, he detailed his attempts to market a cloud API product, a common approach for delivering AI technology. However, feedback indicated that potential customers preferred direct solutions rather than development tools. “What users in the legacy app automation space actually want is to hand me money and receive an automation,” Dunteman explained, highlighting the desire for a more personalized approach similar to hiring a consultant.
With this feedback in mind, Dunteman pivoted to develop Muscle Mem, an AI caching tool. He has opted to step away from the one-off projects typical in Windows automation to build a more scalable solution. Although he declined to elaborate on his decision to abandon the Windows automation concept, Dunteman noted that his new tool is inspired by the problems associated with computer use. Muscle Mem aims to enable agents to delegate repetitive tasks, allowing them to concentrate on more complex reasoning and edge cases.
Dunteman remains optimistic about the future of computer use, referring to it as “the last mile” in AI development. While Pig.dev may have shifted its focus, it’s clear that the challenge of Windows automation is still a relevant issue within the tech community. As more startups explore solutions in this space, the potential for innovation remains vast.
