Google has brought Mechanize co founder Tamay Besiroglu and more than a dozen engineers into DeepMind as it works to improve the software engineering capabilities of its AI models.
The move is structured as a talent acquisition rather than a conventional takeover. Mechanize itself is expected to continue operating under CEO Guive Assadi, while key technical staff move to Google.
The financial terms have not been disclosed. Reports suggest Google had previously discussed a deal worth around $1.5 billion for Mechanize's technology and team. The startup had raised $9.1 million earlier in 2026 at a reported valuation of $500 million.
| Detail | Reported information |
|---|---|
| Company | Mechanize |
| Google division | DeepMind |
| Key hire | Tamay Besiroglu |
| Engineers joining | More than a dozen |
| Main focus | AI software engineering and coding |
| Mechanize funding | $9.1 million |
| Reported earlier valuation | $500 million |
| Reported deal talks | Around $1.5 billion |
| Mechanize CEO | Guive Assadi |
Mechanize talent will focus on model training
The incoming engineers are reportedly concentrating on midtraining, an important stage used to strengthen model capabilities after initial pretraining.
Their work is expected to focus heavily on software engineering tasks.
Coding has become one of the main competitive areas in generative AI, with leading companies trying to make models better at understanding large codebases, finding bugs, completing complex programming tasks and working with development tools.
Mechanize was already focused on improving how AI systems handle software engineering, making its team a logical fit for DeepMind.
Besiroglu also brings experience in AI evaluation. Before Mechanize, he co founded Epoch AI, an organization known for building benchmarks and tracking progress in artificial intelligence.
The deal resembles an acqui hire
Google appears to be using a structure commonly described as an acqui hire.
Instead of purchasing the entire startup outright, a larger company hires key employees and gains access to their expertise while the original business continues separately.
This approach can be simpler than a full acquisition because it may avoid some of the regulatory scrutiny that large mergers face.
Google has used a similar strategy before. It recruited key staff from Windsurf, including former CEO Varun Mohan, while separately advancing its own AI development tools.
That earlier hiring move helped strengthen Google's work on Antigravity, its AI focused development environment.
Coding performance remains an important challenge
The Mechanize hires arrive as Google continues trying to improve Gemini's performance on programming and software engineering tasks.
Recent reports have pointed to delays involving Gemini Pro models because coding performance did not meet internal targets.
That puts additional pressure on DeepMind as competitors continue improving their own coding systems.
AI coding assistants are moving beyond simple code completion. Newer tools are expected to plan larger tasks, edit several files at once, navigate repositories, run tests and complete more autonomous development work.
Improving those abilities requires more than general language performance, which helps explain Google's interest in a team focused specifically on software engineering.
Mechanize is expected to continue separately
The startup is not disappearing entirely as part of the arrangement.
With Guive Assadi taking the CEO role, Mechanize is expected to continue operating while its former co founder and several engineers work inside DeepMind.
That structure resembles other recent AI talent deals where large technology companies recruit key teams without buying every part of the original company.
For Google, the immediate goal appears clear. It wants stronger coding capabilities inside Gemini and its developer tools, and the Mechanize team brings experience directly related to that problem.
Whether the hires significantly improve Google's position will depend on how quickly that expertise translates into better models and development products.
Discussion (0)
Be the first to comment.