From interns to co-founders
An internship at an AI lab focused on autonomous browser agents brought Justin and Nalin together and planted the seed for what would eventually become Scrapybara. Their first YC idea was a web scraper (the “scrapy” in their name), but they always intended to go deeper. They wanted to bring to life the agent concept they’d glimpsed during their internship.
It didn’t take long to realize web scraping was the wrong bet. The space was overcrowded, and AI didn’t offer a meaningful advantage there. So two weeks into YC, they did what every great founder must be willing to do, they pivoted.
The new idea was an AI-powered sales agent. They built it, got customers, but immediately felt the weight of working on something they weren’t excited about.
A last-minute pivot
Around the same time, Anthropic launched Claude’s computer use model. For the first time, an LLM could interact with a computer visually, not solely through code or text. It could actually manipulate a graphical interface like a real user.
They played with it, but kept grinding through SDR demos. “We didn’t believe in it,” Justin says about the AI sales agent. “It was just something we thought we should build, not something we wanted to.”
Things started to unravel, and they hit their low point. Justin came back from a trip sick and burned out. Nalin was alone, taking sales calls for a product they had no conviction in. “We were totally lost,” Justin says. “But then we gave ourselves a day to reset. Two days later, we were building something we knew would work.”
They returned to the idea they’d never let go, building virtual infrastructure for agents that need a desktop. At the time, there was no fast, reliable way to spin up GUI environments for these new models.
With only a week until Demo Day, they pitched the idea to their YC partner, David Lieb, former Google Photos product lead. He told them to go for it.
The all-out sprint to rebuild in days
Most YC companies spend weeks, if not months, prepping for Demo Day. Justin and Nalin had five days. They sprinted to build, shipping a working MVP with a dashboard, API, and SDK in under a week. Within days, they had 500 users. By the end of fundraising, that number had doubled.
They entered Demo Day with a brand-new product and deep conviction that it was the right one. That MVP became the foundation of Scrapybara.
The rapid turnaround and immediate user adoption caught investors’ attention, leading to a flood of interest and an easy seed round raise. “We had speed, a clear gap in the market, and deep knowledge of this space. That combination made a huge difference,” Justin says.
The momentum kept building. As their user base grew, Justin cold-emailed OpenAI asking for early access to their computer use model. They got it, and were ultimately featured as one of the only two companies in the official launch repo. That was a high point. “We were a nobody startup,” Justin said. “But they let us test the model early, we gave tons of feedback, and the next thing we know, we’re featured.”
Like ChatGPT for agent work
Scrapybara is a virtual computer built for AI agents. With a few lines of code, anyone can spin up virtual machines where agents can browse the internet, click buttons, run apps, and go beyond the simple chat functionalities of most agents today.
“We think there’s yet to be a ChatGPT moment for agents,” Justin says. “There’s a way for both the agent and the product to feel approachable to anyone who’s ever used ChatGPT, but hasn’t gone deeper into AI.”
As he and Nalin watched how Scrapybara was being used, they noticed a pattern. People didn’t know which coding tool or model to pick and weren’t sure what they could do with it. So they built Scout, a general agent that works on top of Scrapybara’s infrastructure to help with tasks like coding, research, planning an itinerary, and more. In their launch, Justin and Nalin showed Scout installing dependencies, writing code, and benchmarking Zig and Rust for an N-body simulation in under five minutes. Days later, they showed it deploying a full website from scratch, handling the entire process autonomously.
Building Scout: the most powerful coding agent
Recently, as Justin and Nalin wrapped up development of a new update for Scout, they discovered their most engaged users were using Scout exclusively to code, and not for the 100 other tasks it could handle. So now, they’re focused on making Scout better at coding by integrating with developer environments, strengthening sandboxes with more stable virtual machines, and managing large codebases using subagents, handoffs, indexing, and rules. Their goal is to build a coding agent that solves the core challenges of large-scale development and transforms how developers bring ideas to life.
It’s easy to forget that just weeks before Demo Day, Justin and Nalin scrapped everything to pursue the idea they’d believed in from the start. That five-day sprint became the foundation of Scrapybara and Scout, marking a bet on the future of agents designed to bridge the distance between idea and reality.