# Google weighs $1.5 billion-plus deal for Mechanize staff and AI technology > Source: > Published: 2026-08-05 16:58:59+00:00 Google is in talks on a deal worth more than $1.5 billion with Mechanize, the San Francisco AI training startup founded by [Tamay Besiroglu (@tamaybes)](https://x.com/tamaybes), [Matthew Barnett (@MatthewJBar)](https://x.com/MatthewJBar), and [Ege Erdil (@EgeErdil2)](https://x.com/EgeErdil2), according to [Business Insider](https://www.businessinsider.com/google-mechanize-deal-talent-tech-ai-coding-2026-8). The transaction under discussion is narrower than an acquisition. Google would hire some Mechanize employees and take a non-exclusive license to its technology, Business Insider reported, citing four people familiar with the talks. The recruits would work on AI model evaluation and development. Terms remain under negotiation and could change. That distinction matters because the $1.5 billion-plus figure is the reported value of the overall arrangement, rather than a purchase price for Mechanize. [Deedy Das (@deedydas)](https://x.com/deedydas) described the talks [in a post on X](https://x.com/deedydas/status/2085037385927291067) as an acquisition, but the underlying report describes a licensing and hiring package that would leave Mechanize outside Google's ownership. Mechanize was launched on April 17th, 2025, by Besiroglu, Barnett, and Erdil, three researchers who had worked at Epoch AI. Besiroglu, now Mechanize's CEO, co-founded Epoch, an AI research institute known for measuring model capabilities, compute trends, and the economic effects of automation. Mechanize turned that evaluation work into a commercial proposition: building simulated workplaces where AI agents can practice tasks and receive automated scores. The founders set a wider goal from the beginning. In [Mechanize's launch announcement](https://www.mechanize.work/announcing-mechanize-inc/), they said they wanted to create the environments, benchmarks, and training data required to automate work across the economy. Mechanize started with software engineering, a field where tasks can be executed inside controlled computer environments and graded against tests or other verifiable outcomes. ### The scarce input behind coding agents Mechanize's product addresses a bottleneck that has grown as AI labs push coding systems beyond generating snippets and toward completing longer assignments. Models need environments containing a prompt, a working codebase, and a grader capable of determining whether the agent completed the job correctly. Those environments can then be used for reinforcement learning or model evaluations. Mechanize says a single engineer owns each task from idea through grading and quality assurance. A typical task takes about one week, and the bulk of the work goes into making its grader fair, deterministic, and difficult for a model to exploit, according to the startup's [description of its production process](https://www.mechanize.work/what-working-here-is-like/). Mechanize currently lists a team of around 35 people. That operating model helps explain why Google is considering a deal centered on people as well as intellectual property. Much of Mechanize's value sits in the judgment required to identify genuine weaknesses in frontier coding models, construct realistic assignments around them, and distinguish model failures from flaws in the test itself. A technology license would give Google access to completed work. Hiring Mechanize employees would bring in the people who know how to keep producing it as models improve. The proposed transaction would also give Mechanize's founders a rapid jump in financial scale. Business Insider reported that Mechanize said earlier in 2026 that it had raised $9.1 million at a $500 million valuation. Its backers include former GitHub CEO Nat Friedman, former Y Combinator president Daniel Gross, Stripe CEO Patrick Collison, podcaster Dwarkesh Patel, Sholto Douglas, and Marcus Abramovitch. The startup's current website also names Adam D'Angelo, Marco Mascorro, Devendra Chaplot, and Alex Atallah among its investors. The reported $1.5 billion-plus package should not be read as a new valuation without knowing how the money would be divided among licensing fees, employee compensation, founders, and investors. The structure could produce a large payout while leaving a reduced Mechanize operating independently. ### Google returns to the reverse-acquihire playbook Google used a similar structure in July 2025 when it hired Windsurf co-founders Varun Mohan and Douglas Chen, brought several colleagues into Google DeepMind, and obtained a non-exclusive license to Windsurf technology. That package was worth about $2.4 billion, according to reporting at the time. Mohan now leads work on Google Antigravity, its agentic coding platform. The Mechanize talks would extend that strategy deeper into the AI coding supply chain. Windsurf brought Google a team that had built a developer-facing coding product. Mechanize would supply people and infrastructure used to train and evaluate the underlying models. These arrangements let technology companies secure founders, researchers, and intellectual property without completing a conventional takeover. They also leave difficult questions for the startup employees and investors who remain behind. A non-exclusive license preserves the company's legal independence, while the departure of key personnel can determine whether that independence has much commercial value. For Besiroglu and his co-founders, the talks arrive roughly 16 months after they converted years of research on AI capabilities and automation into a startup. For Google, the proposed price reflects how quickly high-quality training environments have become strategic assets in the contest to build coding agents that can complete real software work rather than simply generate code on command.