Hassabis Says AI Job Cuts Are the Wrong Strategy

Demis Hassabis is not trying to soften Google DeepMind’s technical ambition in his interview with Wired. Quite the opposite: he describes Gemini 3.5 Flash as a model that can translate large codebases, find and fix bugs, and write highly complex software. But he draws a conclusion that does not fit the simplest corporate script: if AI raises productivity, companies should do more and aim higher, not simply reduce headcount.

That distinction matters. The AI coding debate often gets stuck between two caricatures: one where models immediately replace developers, and another where they are only smarter autocomplete. Hassabis is pointing at a third zone. According to the source summary, he rejects the confidence with which parts of the industry talk about mass replacement of software engineers, framing that certainty as a lack of imagination and a weak reading of what is actually happening.

Google’s context is obvious. Google DeepMind wants Gemini to become infrastructure for serious work, not just a presentation chatbot. The official Gemini project page already positions it as a central model layer, while the article frames Gemini 3.5 Flash as a practical engine for code. If a model can translate legacy systems, locate defects, and generate parts of an operating system, the question is no longer whether it can assist developers. The sharper question is whether organizations can change work quickly enough to avoid turning that assistance into a cost-cutting spreadsheet.

That is where Antigravity enters the story. Google revealed the coding tool at its annual developer event, and the precise product shape is less important than the direction of travel. Whether it functions as a standalone environment, an agentic layer, or a tighter integration into existing workflows, Google is trying to move AI into the daily mechanics of software development. In that setting, value is not measured by a demo clip. It is measured by how safely teams can change existing code, how many bugs are caught, and how quickly the system returns to an understandable state.

Hassabis’ argument is not sentimental. It is strategic. If a company treats AI only as a wage-reduction mechanism, it may look efficient in the short term while losing the people who understand architecture, risk, and the consequences of machine-generated code. In software, speed without responsibility can become technical debt with better branding.

That also explains why the broader coding market matters. Anthropic’s coding tools and OpenAI’s platform systems for working with code, including its code interpreter tooling, show that AI coding is no longer a side feature. It is one of the most important commercial tests for the new model generation: can these systems help inside real, messy, long-lived software projects?

The most interesting part of Hassabis’ position is therefore not a moral defense of existing jobs. It is the claim that layoffs may signal weak management. If a tool lets a team translate an old system, fix defects faster, or launch a product that was previously too expensive, cutting people is the shallowest possible interpretation of productivity. The smarter test will be what companies do with the new capacity after the first model demo stops feeling magical.