Why Google Might Spend $1.5 B on a 35‑Person Startup to Accelerate AI Coding

Google is in talks to pay more than $1.5 billion to acquire San Francisco startup Mechanize and its technology, aiming to boost its AI‑coding capabilities; four insiders say the negotiation is still ongoing and terms may change.

Why 35 People Cost $1.5 B

Mechanize, founded in 2025 by three former Epoch AI researchers, now has 35‑50 staff members who earn roughly $400,000 each per year and complete about 1,003 tasks weekly at an average cost of $8,000 per task. Its mission is to automate all work, starting with software because code can be objectively scored.

The company creates training environments where an AI agent receives a prompt, a runnable codebase, and a scorer that judges whether the task is completed. Each task is handcrafted by an engineer over a week to ensure fairness, consistency, and to prevent model cheating; the resulting data feed reinforcement‑learning pipelines and model evaluation.

Google’s Acquisition Strategy

Google’s plan is not a full takeover but a hiring and non‑exclusive licensing arrangement. The hired staff will evaluate and develop Google’s models, while the licensed evaluation framework will help identify weaknesses in coding models without waiting for the next Gemini generation.

Google has used similar tactics before, such as hiring the founders of Windsurf and licensing its coding tools for about $2.4 billion in 2025, and re‑hiring Character AI co‑founder Noam Shazeer. This structure avoids the antitrust scrutiny that a complete acquisition would trigger.

Market Context and Timing

The deal arrives as Google’s AI division faces setbacks: DeepMind CEO Demis Hassabis stepped down, chief scientist Jeff Dean left, and Gemini has repeatedly lagged behind OpenAI’s Codex and Anthropic’s Claude Code. Competitors are building their own coding agents, making Mechanize’s evaluation suite a way to narrow the gap quickly.

Analysts note that the $1.5 billion figure likely covers licensing fees, salaries, and founder payouts, and that Mechanize will remain an independent entity. The core scarcity in AI programming is now talent that can create robust test cases rather than the models themselves.