Soham Parekh aced interviews at top AI startups without solving a single LeetCode problem. And yet, every company hired him.
By all accounts, Soham Parekh is a capable engineer. Over the past two years, he held overlapping roles at some of Silicon Valley’s most closely watched AI startups including Antimetal, Pally AI, and Sync Labs. Each company believed they had hired a focused, high-velocity contributor. Each eventually discovered they weren’t the only ones.
Parekh wasn’t underperforming. He was delivering. “There were two occasions where the contractor bailed,” he recalled in a recent podcast interview, referencing his time at Antimatter. “I wrote every single inch of the code.”
He also didn’t hide how he got through the door. “Most of these companies didn’t have LeetCode-style interviews,” he said. “If they did, I would’ve bombed them. I don’t practice LeetCode that much.”
The Culture That Made It Work
The AI startup ecosystem in 2025 runs lean, remote, and async. Code output is prioritized over presence. If work is delivered and builds don’t break, few founders ask further questions. Parekh knew this and leaned into it.
He described himself as a “serial non-sleeper,” someone who works until a task is done—regardless of how many other tasks he might also be juggling. “I just worked until I got something done for the day,” he said.
That approach aligns with what many fast-moving teams want: engineers who are independent, self-managing, and always shipping. But in an environment flooded with powerful AI developer tools, it’s getting harder to know who or what is producing that output.
Parekh said his multi-job run started before tools like Copilot became ubiquitous. “Some of these companies I worked at were before the Copilot boom,” he said. “There were no AI-assisted programming tools.”
That changed by 2023–24, with widespread adoption of GitHub Copilot, Cursor, Claude, and Devon. These tools didn’t give Parekh an unfair advantage; they simply helped him move faster. “Obviously, having more AI-assisted tooling definitely helped,” he said. “But it did not amount to me working on more jobs… I was just trying to get things done.”
As he put it, the work wasn’t being offloaded or outsourced, but it was being accelerated. “I don’t do anything outside coding. That is very true.”
Why LeetCode Didn’t Come Up
Parekh made no secret of his avoidance of algorithmic interview prep. “If they had LeetCode-style interviews, I would’ve bombed them,” he repeated. “I don’t practice LeetCode that much.”
That’s not unusual in the current startup landscape. Many AI-first companies have moved away from traditional DSA (data structures and algorithms) interviews in favor of async take-homes, GitHub activity reviews, or short trial projects. These formats often reward developers who can ship fast, write clean code, and get up to speed quickly.
But as generative AI tools have become more sophisticated, those same formats have become more susceptible to automation. Boilerplate code, integration tasks, and unit tests can now be scaffolded in minutes. A polished take-home doesn’t always mean an engineer was deeply engaged. It might mean they knew how to prompt the right LLM.
Still, in Parekh’s case, there was no faking. “I had no junior devs under me,” he said. “Everything I built, I built myself.”
A System That Missed the Signals
Founders who hired Parekh didn’t complain about his output. Some, like Suhail Doshi of Playground AI, only flagged issues a week in. Others didn’t notice anything until much later. The Verge reported that several described him as technically strong, fast, and collaborative.
But the issue wasn’t productivity. It was visibility. In a system built around speed, few people questioned where that speed was coming from or how it was being sustained across multiple teams.
“I think startups usually just care more about whether you can work,” Parekh said. “And I do like what I do.”
Despite the mounting scrutiny, Soham Parekh appeared on the Technology Business Programming Network to share his side of the story, clarifying that he had been juggling multiple jobs since 2022 without relying on AI assistants or delegating work to junior engineers. He insisted the experience had made him a better programmer, though it came at a personal cost.
While some critics labeled him a scammer, Parekh seemed ready to turn the controversy into opportunity. He briefly announced a new role at Darwin Studios, an AI video remixing startup only to delete the post shortly after, as did Darwin’s founder and CEO, Sanjit Juneja. In a statement shared with TechCrunch, Juneja stood by the decision, saying, “Soham is an incredibly talented engineer and we believe in his abilities to help bring our products to market.”
