The Idea:
Peter Albert took a job at Meta with an unusual side quest. He wasn’t there for the salary or the title. He was there to find a co-founder.
He landed in Meta’s AI ranks, on the core team behind Llama 2, one of the most important open-source language models ever released. Nearby was Fryderyk “Fryd” Wiatrowski, a Warsaw-raised Oxford maths and computer science graduate who’d already done a stint in high-frequency trading before joining Meta.
The pair started spending their evenings building AI agents. The frustration was obvious to anyone working in AI in 2023. Language models could write an essay in seconds, but they couldn’t actually do anything. Every piece of real office work still meant a human clicking through a dozen different apps, one tab at a time.
Their bet: give AI the digital equivalent of arms and legs, and it could handle the tedious work that ate everyone’s week.
Six months into his full-time role, Wiatrowski walked out. Albert went with him.
This was the beginning of Viktor.
The Execution:
- August 2023: Wiatrowski and Albert founded Zeta Labs, the lab that would eventually build Viktor, with one goal: teach AI models to operate software the way humans do.
- June 2024: Zeta Labs raised a $2.9M pre-seed led by Daniel Gross and Nat Friedman, with ElevenLabs founder Mati Staniszewski also chipping in. It launched JACE, an agent that controlled a web browser, and to prove the point, had it set up a company on its own.
- 2024 to 2025: Reality bit. Browser agents at the time managed three to five steps at roughly 60% reliability, far too flaky for real work. The team retreated to the inbox and relaunched Jace as a Gmail assistant at up to $65 a month. It found paying customers, but the agent ran on every single incoming email and the margins never worked. Wiatrowski later admitted the inbox was the wrong home for an AI agent.
- February 2026: Two and a half years in, they launched Viktor, an “AI employee” living inside Slack with its own cloud computer, plugged into 3,000+ tools. By March it was doing around $1.5M in annualised revenue. Not every early shift went smoothly: Viktor once reacted to a layoff announcement with a skull emoji in a public channel.
- May 2026: Ten weeks after launch, Viktor hit a $15M revenue run rate across 2,000+ organisations. Accel had tracked the founders for two years but found their earlier products too niche. This time it flew to Warsaw within weeks and led a $75M Series A, joined by Slack co-founders Stewart Butterfield and Cal Henderson. It was reportedly the largest Series A ever for a Polish-founded company.
- June 2026: Viktor landed in Microsoft Teams and crossed $20M ARR the same week, with zero salespeople. At that point it had around 2,900 paying customers spending an average of $400 a month.
- August 2026: Viktor passed $30M ARR with just 35 employees. Net dollar retention in the first 30 days ran at 350%: a customer who started at $1 was paying $3.50 a month later.
- Today: Viktor says it’s used by 71,000+ teams, connects to 3,200+ tools, and is opening a New York office to chase US enterprise customers.
The lesson? Zeta Labs never had the wrong idea. It had the wrong address. The same agent failed four times in ten inside a browser, then burned margin on every email inside an inbox, before landing in Slack, where a ten-minute wait feels fast because you’ve asked a colleague rather than clicked a button. Two and a half years of “too niche”, then $30M ARR in six months. Before you kill the product, try moving it.