Cyberdesk

We met 2 years ago, connecting over our shared experiences at Rice, and our love of building things. We both ended up living at the same Hacker House in Menlo Park, called Incepto! We started Cyberdesk with the realization that it’s really hard to build a great computer use agent, especially when it comes to reliability, speed, and cost. We both lived in a 9×9 Airbnb room with a tiny bed and a single desk (we call it the Cyberdesk lol) and got Cyberdesk off the ground, building the MVP in a 6 days and getting a paid customer the next day through a cold email.

Cyberdesk was built to solve all three of the core problems with off the shelf computer use models: speed, reliability, and cost. We’re excited to deploy this architecture to the world and become the best developer framework for computer use.

From the moment we started working together, we knew we wanted to go through YC together. It was a little over a month before the deadline for Summer 2025 came out when we started working together, so we got to work! We launched, got paid customers, and built out our infrastructure fast. A month later, we interviewed in person and got the call from Jon Xu the same night that we got in! It was one of the happiest days of our lives.

We started working together in late March (2025), with a shared goal of solving deep developer problems around computer use. We started off with an idea to help developers spin up virtual desktops for computer use agents in a single line of code. We shipped an MVP in 5 days and got a paid customer the next day through a cold email. For the next month, we kept shipping and looking for customers, and applied to YC with this idea. We found the product had interest from hobbyist developers, but for companies, there was much less pull because companies often already have VM infrastructure they are happy with.

So we made the call to pivot. We talked to over 100 developers in the computer use and robotic process automation space, and heard the same thing over and over: automating Windows desktops is an important yet extremely tedious process. You create a script once, then it breaks every time something changes in the app or a popup happens. Computer use agents were meant to solve that pain but even they were slow, costly, and non deterministic. That’s how we came up with the idea for what Cyberdesk is today - a layer on top of computer use agents that helps it memorize tasks it’s done before and repeat it consistently, 3 - 5x faster, and cheaply, while still falling back to computer use during popups and anomalies. We built an MVP in a week, and got our first pilot a week later.

Since then we’ve been rapidly building out the product and closing more deals with startups and companies in healthcare, construction, accounting, and more, all of them using our agent to automate manual computer work they’ve struggled to automate before.

For a developer team to automate legacy desktop apps with poor APIs, they’ve often resorted to building brittle automation scripts (either in Python or using a legacy RPA tool like UiPath) or hiring unreliable remote workers to click on remote desktops. As they need to support more and more software and versions of that software, the development and maintenance times exponentially increases.

Every company in the world has repetitive manual desktop work that is bogging them down. On top of that, companies building AI agents are increasingly having to interface with desktop apps to perform full human workflows. This problem is pervasive and we’re excited to solve it.

Traditional robotic process automation is a $6 billion industry, yet it only services a tiny fraction of desktop tasks that can now be automated by our agent. In the long term, our agent will automate every manual computer task.