Demis Hassabis Believes AI Job Reductions Are Unwise

Demis Hassabis, the CEO of Google DeepMind, is eager to discuss the programming capabilities of his company’s latest model, Gemini 3.5 Flash. This model has been trained to tackle sophisticated coding tasks: translating extensive codebases between languages, identifying and resolving bugs hidden in complex code, and even creating complete operating systems from scratch.

However, Hassabis is not convinced that this indicates a grim future for software developers. “I have no idea why people are spreading such definitive statements,” Hassabis tells WIRED ahead of the new model unveiling at today’s Google I/O event.

“Perhaps there’s an ulterior motive behind these claims; perhaps for fundraising or something similar,” Hassabis suggests. “From my perspective, and that of DeepMind and Google, if engineers become three or four times more productive, we simply want to [aim to] accomplish three or four times more.”

The remarkable coding proficiency of the latest models has instigated a widespread concern that AI could soon render programming jobs and other white-collar positions obsolete. Executives at various AI firms have forecast broad job displacement, while several notable tech companies, such as Amazon, Salesforce, and Block, attribute recent layoffs to the increasing use of AI.

Hassabis believes that Alphabet, which governs several companies in addition to Google, might be strategically positioned to leverage a transformation in software productivity. “I have countless ideas, from drug discovery in labs to game design,” he states. “I’d love to have some additional engineers available to pursue those kinds of projects.”

Hassabis argues that companies seeking to replace developers with AI may be making a significant error. “I think it’s a lack of vision—and a misunderstanding of what is genuinely going to happen,” he remarks.

At its annual developer event, Google showcased a variety of AI advancements. With a coding tool named Antigravity, Gemini 3.5 Flash delivers cutting-edge coding and reasoning abilities but is noted to be faster and more economical than its competitors, according to Google. Gemini 3.5 Pro, a more robust version of its flagship model, is set to launch next month.

The company must enhance its AI coding capabilities, which have become a key and profitable application for the newest AI models. According to a 2025 Stack Overflow survey, Anthropic and OpenAI are leading developer adoption with their tools, Claude and Codex.

Google also demonstrated an assistant called Spark that operates within Google Cloud and has access to its applications. The design aims to be safer than alternatives like OpenClaw by limiting access to personal data, as stated by Google.

Other demonstrations included a version of Android featuring an integrated AI agent and an updated Google Search that employs agentic coding to dynamically generate a site or app in response to a search query.

AI coding has captured the attention of the AI community in recent months, even sparking hopes that models could eventually rewrite their own code in a cycle of self-improvement. Hassabis acknowledges this possibility but is skeptical that it will lead to superhuman-level AI in the short term.

Advancements in other scientific fields might necessitate AI models acquiring a more profound understanding of the physical world and the ability to conduct experiments within it, he notes.

Even in the seemingly completed domain of coding, Hassabis points out that it is significant that AI has yet to create a blockbuster app or video game without human assistance. “I think something is lacking,” he concludes.