Google has reportedly reached late-stage negotiations to take a technology licence from Mechanize and hire part of its team, in an arrangement Business Insider values at more than $1.5 billion. Neither company has confirmed a deal, and Business Insider’s own account notes the terms could still change before anything closes. The mechanism matters as much as the number: this looks like a license-and-hire structure, not a straight acquisition.
Mechanize, founded in 2025 by Tamay Besiroglu, Matthew Barnett, and Ege Erdil, does not build a chatbot or a general coding assistant. It supplies the simulated settings, the scoring suites and the training corpora used to grade and train AI agents on complex, multi-step software work, with a stated goal of automating software engineering first and other knowledge work after that. The company has said existing coding agents handle narrow tasks well but still struggle with the long-context, ongoing work that real software projects require.
That focus is the more interesting fact here than the price tag. When a frontier lab shops for a coding startup to acquire, the target is usually a product. When the target instead builds the scaffolding that measures what an agent can do and generates the data it trains on, the shopping signals something different: the bottleneck on agent capability has moved from the model itself to the environments and evaluations used to shape it.
The pattern also is not new for Google. In July 2025, the company paid roughly $2.4 billion to license technology from Windsurf and bring across chief executive Varun Mohan, fellow founder Douglas Chen and a number of researchers, who now work inside Google DeepMind on agentic coding tied to Gemini, according to Reuters. A Mechanize deal built the same way, license the technology, absorb some of the team, would repeat that structure rather than introduce a new one. Structuring deals this way lets Google add capability and talent without the merger scrutiny a full acquisition can invite.
The valuation jump, if the reported figure holds, is the part still unexplained. Mechanize took in $9.1 million during April, a round that valued it at $500 million once the money was in. A price above $1.5 billion four months later would mark roughly a threefold increase, and neither Business Insider’s reporting nor Mechanize’s public statements account for what changed in the interim.
Google’s interest arrives inside a coding race that has already reshaped how the major labs describe their own engineering. Anthropic has built Claude Code into a product that navigates repositories, edits files, and runs tests rather than just autocompleting lines. OpenAI is developing Codex on similar lines, and Meta recently entered with its own agent, Muse Code. Google introduced Gemini 3.5 Flash and agentic coding tools at I/O this year, followed by Gemini 3.7 Flash, aimed specifically at coding and agent workloads.
None of this is confirmed. Google and Mechanize both declined to comment, and Business Insider’s report is explicit that details could still shift or the talks could collapse before any signing. What deal value survives, which Mechanize staff would move, and how the technology would slot into Gemini and DeepMind’s existing coding work remain unstated.
Teams evaluating coding agents this quarter should treat eval and environment infrastructure, not just model quality, as the thing to watch: if Google closes this deal, expect rivals to bid for the same category of infrastructure rather than for more chatbot talent.
Reported by gHacks on 26 August 2026, citing Business Insider’s original reporting.
