Google has completed a deal worth more than $1.5 billion for talent and technology from AI-agent startup Mechanize, bringing more than a dozen employees into DeepMind, including co-founder and former chief executive Tamay Besiroglu. Reporting on the transaction says the team will work on model development, evaluation and the midtraining layer used to improve capabilities such as coding.
The structure matters more than the headline price. When negotiations emerged in August, the reported deal was a non-exclusive licence to Mechanize technology combined with the hiring of staff. The corporate entity was outside the transaction. September's completion report describes the same basic structure: Google gets people and licensed access while Mechanize remains a separate company under new chief executive Guive Assadi.
Mechanize had raised $9.1 million and had been valued at roughly $500 million, but that valuation priced equity in the company. Google's $1.5 billion-plus package covers an undisclosed mix of employment compensation, retention, licence rights and other deal terms while Mechanize equity remains outside the transaction. Treating the difference as a threefold takeover premium compares different assets.
The scarce asset is the feedback loop for training agents
Mechanize has focused on simulated environments, evaluation benchmarks and grading systems for AI agents. Those tools sit inside the loop used to test whether an agent can complete difficult software tasks and to turn task performance into training signals. DeepMind is buying into that feedback loop: people and tooling that can accelerate how models are tested and improved.
The non-exclusive licence gives Google technology access while leaving corporate ownership and future equity value with Mechanize's shareholders. Licence duration, field of use and breadth remain private, leaving the durability of Google's access rights unpriced in the headline figure.
The price reveals where Google sees scarcity. More than $1.5 billion for a team measured in dozens plus technology access sits far outside normal hiring economics. Dividing the consideration by headcount would be meaningless because a material share may belong to the licence or other rights. The useful signal is capital allocation: Alphabet is willing to spend at acquisition scale to shorten the path to better agent training while leaving the company independent.
The remaining valuation question is the consideration waterfall. If most of the $1.5 billion-plus is long-dated retention and employment compensation, the headline number is primarily the cost of keeping scarce people. If a large portion is upfront licence consideration for durable technology rights, the economic asset is different. The public record leaves those components bundled together.
For Alphabet shareholders, the deal is best read as acquisition-scale spending on scarce model-development capability with a narrower legal perimeter than an acquisition. Mechanize stays independent; DeepMind gets the team and licensed technology. The consideration waterfall will show whether most of the $1.5 billion-plus paid for durable technology rights or scarce people. Either way, Google chose access over corporate control.