TLDRocket
Sign in

How Mechanize went from a $9M seed to $1.5B Google acquisition talks in just 100 days

Tech Funding News Sofia Chesnokova

Google's reportedly offering $1.5B for Mechanize's tech and team, just 103 days after the startup raised a $9M seed. It's Google's third reverse acqui-hire in two years, and this one turns a Twitter villain into a billionaire's target.

Tamay Besiroglu spent April 2025 getting called a traitor on X. By April 2026, he'd raised $9.1 million for Mechanize at a $500 million valuation. Now, barely 103 days later, Google is reportedly weighing a deal worth more than $1.5 billion to license the startup's tech and absorb its engineering team. That's roughly triple the valuation investors agreed to just months earlier, which tells you something about how fast the ground is shifting under AI infrastructure right now.

Besiroglu, along with co-founders Matthew Barnett and Ege Erdil, built Mechanize after leaving Epoch AI, the nonprofit Besiroglu himself co-founded in 2022 to keep tabs on advanced AI systems. Going from watchdog to automation vendor didn't sit well with everyone. PauseAI's Holly Elmore publicly accused him of betraying the mission the moment he announced the pivot. But the pivot worked, at least commercially. Mechanize builds simulated coding environments and evaluation systems — think a test that asks an agent to build a working Game Boy Advance emulator in 24 hours, something you can't just pass or fail like a typical benchmark. That kind of realistic, multi-step evaluation is exactly what frontier labs need to train coding agents that don't just game simplified tests.

The RL-environment space is getting crowded fast. Prime Intellect has an open-source hub with over 2,500 community environments. Daytona sells sandboxes for AI-generated code. Mercor already snapped up a smaller player, Deeptune, back in July. Mechanize took the opposite approach from most of these companies: instead of building something broad, it narrowed in on a small number of high-quality coding evaluations aimed squarely at frontier labs. That focus is likely the whole pitch to Google, which has been playing catch-up to Anthropic and OpenAI on developer coding tools ever since Claude Code and Codex got there first. Coding remains one of the rare corners of AI that actually generates real revenue, so closing that gap matters more than most.

This isn't Google's first move of this kind, either. It brought back Character AI's Noam Shazeer in 2024 while licensing that startup's tech, then paid $2.4 billion for Windsurf's technology and poached CEO Varun Mohan in 2025 after OpenAI's own $3 billion offer collapsed under Microsoft's objections. Microsoft and Amazon have run comparable plays with Inflection AI and Adept. None of these deals trigger a standard merger review, which regulators have noticed — they've said they'll start scrutinizing acquisitions structured specifically to dodge that process.

A year ago, Mechanize was a footnote in an online argument about AI safety. Now it's apparently worth $1.5 billion to one of the largest companies on earth. Whether that number becomes the new floor for RL-environment startups, or just reflects what Google was willing to pay to move first, is anyone's guess for now.

My take

Structuring these deals as licensing-plus-hiring instead of outright acquisitions is a legal workaround dressed up as innovation, and everyone involved knows it. Big Tech gets the talent and the tech without the antitrust paperwork, regulators mutter about closing the loophole, and nothing actually changes. If watchdogs are serious about scrutinizing these reverse acqui-hires, Mechanize's deal is a perfect test case — but don't bet on anyone actually blocking it.

Read more about this at: Tech Funding News

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.