Google Eyes $1.5bn Mechanize Deal to Close AI Coding Gap

Alphabet is negotiating a licence-and-hire arrangement with San Francisco-based Mechanize, a startup that builds evaluation environments for coding agents, as it seeks to catch up with rivals OpenAI and Anthropic.

The Mechanize Talks: What Google Wants and Why

Alphabet is in advanced talks with Mechanize, a San Francisco startup barely a year old and counting just 35 employees, in a deal valued at more than $1.5 billion. Rather than a conventional acquisition, the structure would see Google hire some of Mechanize’s staff and take a non-exclusive licence to its technology, according to Business Insider. Both companies have declined to comment, and the terms could still shift.

Mechanize was founded in April 2025 by three former researchers from Epoch AI, an institute that measures model capabilities. Its chief executive, Tamay Besiroglu, co-founded Epoch. The startup had already raised $9.1 million at a $500 million valuation from backers including former GitHub chief Nat Friedman, Stripe’s Patrick Collison and podcaster Dwarkesh Patel. Its blunt mission — to automate every job — started with software because code can be graded objectively.

That grading is the company’s real product: bespoke training environments where coding agents receive a prompt, a working codebase and a fair, cheat-resistant grader that decides if the agent succeeded. Each task takes an engineer about a week to build, and the bulk of the effort goes into making the evaluation consistent and difficult to game. Google would get a licence to this finished work and, crucially, the people who keep it evolving as models improve.

Google’s Repeated Playbook: Licence, Hire, Sidestep Antitrust

The Antitrust-Acquihire Pattern

The Mechanize talks extend a well-established Google playbook. In July 2025, it hired the founders of coding tool Windsurf and took a non-exclusive licence to their technology in a package reported at about $2.4 billion. That deal placed Windsurf’s CEO Varun Mohan at the head of Google’s Antigravity coding platform. A year earlier, it rehired Character AI co-founder Noam Shazeer and paid for rights to that startup’s work. The structure lets Google absorb talent and IP while avoiding the full antitrust scrutiny of a straightforward acquisition.

Why Coding Evaluation Is Worth $1.5bn

Mechanize’s value lies not in a consumer product but in the infrastructure that trains frontier models. As coding AI shifts from models to the tools around them, the scarce resource is no longer the model itself but the ability to build rigorous tests that reveal where those models fail. Google’s flagship Gemini has repeatedly slipped in coding benchmarks while OpenAI’s Codex and Anthropic’s Claude Code have pulled ahead. By buying the machinery that grades and trains coding agents, Google aims to close the gap without waiting for the next Gemini iteration.

The reported $1.5 billion figure is not a purchase price or a fresh valuation but the total value of the arrangement — split across licensing fees, salaries, and payouts to founders and investors. Mechanize would remain an independent company, albeit one heavily intertwined with Google’s development pipeline.

A Tough Week for Google’s AI Ambitions

The talks surfaced amid visible strain in Google’s AI unit. Days earlier, DeepMind boss Demis Hassabis stepped back and chief scientist Jeff Dean left to start his own company, triggering a roughly 4% drop in Alphabet shares. The departures underscored talent instability and delays that have already raised questions about Google’s ability to keep pace in the fast-moving coding-agent race.

What the Deal Means for Tech Leaders and Investors

For technology leaders:

Organizations building internal AI coding tools should note the premium Google is placing on evaluation rigour, not just model capability. Investing in proprietary testing and grading infrastructure may soon be as critical as model development.
The licence-and-hire blueprint bypasses M&A thresholds but can still transfer key talent and IP. Companies with coveted AI teams should expect similar overtures from cash-rich rivals, potentially altering the competitive landscape without formal takeovers.

For investors:

The deal structure could reset expectations for AI startup exits, demonstrating that non-acquisition arrangements can command valuations in the billions, though exact returns for early backers depend on how the $1.5 billion is apportioned.
Startups focused on evaluation and training infrastructure — a previously overlooked niche — may now command outsized attention and valuations as the AI coding race intensifies.

Risk & Opportunity Assessment

Commercial Risk	Medium	Google’s AI coding tools face stiff competition from OpenAI and Anthropic, and a failure to integrate Mechanize’s technology quickly could see Gemini continue to lose developer mindshare.
Competitive Risk	High	Rivals are building their own coding agents, and Google’s repeated slippage in coding benchmarks has left it chasing. The Mechanize deal is a direct response to this competitive gap.
Regulatory Risk	Low	The licence-and-hire structure is specifically designed to avoid the antitrust scrutiny of a full acquisition, though regulators could still examine the arrangement if it raises market-power concerns.
Reputation Risk	Medium	The departure of senior DeepMind leaders and the perception that Google is paying heavily to catch up — while Apple pays Google to use Gemini — could reinforce a narrative of a diminished AI pioneer.
Technology Disruption	Transformational	AI that can reliably write and test code threatens to reshape software development entirely. Google’s investment underscores that the bottleneck has moved from model size to evaluation and training infrastructure.
Commercial Opportunity	High	If the deal closes, Google gains a proprietary evaluation engine and the talent to continuously improve it, potentially accelerating Gemini’s coding capabilities and reclaiming developer share from rivals.