Google DeepMind Acqui-Hires AI Coding Startup Mechanize Team to Boost Models
Google has completed an acqui-hire of San Francisco-based AI programming startup Mechanize, bringing co-founder Tamay Besiroglu and over a dozen engineers into Google DeepMind. The incoming team will focus on LLM mid-training to significantly upgrade the coding and software engineering capabilities of Google's AI models. The talent influx aims to help Google overcome recent performance gaps in AI software engineering, allowing it to compete more aggressively against rivals like OpenAI and Anthropic. Furthermore, utilizing an acqui-hire structure enables tech giants to secure critical talent and technology licensing without triggering lengthy antitrust reviews by regulators such as the FTC or CMA. Reports indicate Google previously negotiated a deal valued at up to $1.5 billion for Mechanize's talent and technology access, following Mechanize's prior $9.1 million funding round at a $500 million valuation. Mechanize co-founder Tamay Besiroglu also co-founded Epoch AI, bringing extensive domain expertise in AI performance benchmarking.
## BACKGROUND
Mid-training is a crucial bridge in large language model development between pre-training and post-training, utilizing techniques like learning rate annealing and domain-specific dataset curation to refine capabilities. Epoch AI is a prominent research organization dedicated to tracking AI development trajectories and establishing rigorous benchmarks.