Demis Hassabis, The CEO of Google DeepMind is eager to talk about the encoding capabilities of his company’s latest model, Gemini 3.5 Flash. The model was trained to perform intricate agent-based coding tasks: translating gigantic codebases from one language to another; find and fix errors lurking deep in intricate code; and even write entire operating systems from scratch.
But Hassabis doesn’t think this spells doom for software developers. “I have no idea why people are talking about this with such certainty,” Hassabis tells WIRED before unveiling the modern model at today’s Google I/O event.
“Maybe there is an ulterior motive for spreading these messages; raising money or whatever,” Hassabis says. “From my point of view, from the point of view of DeepMind and Google, if engineers become three or four times more productive, then it will simply [want to] do three or four times as many things.”
The striking coding capabilities of the latest models have sparked widespread fear that artificial intelligence may be on the verge of eliminating programming and other white-collar jobs. Executives at some AI companies have predicted widespread job relocation, while some leading technology companies, including Amazon, SalesforceAND Blockthey blame the employ of artificial intelligence for recent layoffs.
Hassabis believes Alphabet, which oversees several companies in addition to Google, could be well-positioned to take advantage of the software productivity revolution. “I have a million ideas, from laboratory drug discovery to game design,” he says. “I’d like to have some engineers available to do this kind of thing.”
Hassabis says companies looking to replace programmers with artificial intelligence could be making a massive mistake. “I think it’s a lack of imagination and a lack of understanding of what’s really going to happen,” he says.
Google showed off a slew of artificial intelligence solutions at its annual developer event. With an encoding tool called Antigravity, Gemini 3.5 Flash offers pioneering encoding and inference capabilities, but is faster and cheaper than competitors’ products, Google says. Gemini 3.5 Pro, a modern, more powerful version of its flagship model, will debut next month.
The company needs to catch up when it comes to AI coding, which has emerged as a key and lucrative application for the latest AI models. Adopting leading Anthropic and OpenAI developers with the right tools, Claude and Codex, by 2025 Stack Overflow Survey.
The company also demonstrated an agent assistant called Spark that lives in Google’s cloud and has access to its applications. The project is intended to be more secure than something like OpenClaw because it has circumscribed access to personal data, Google says.
Other agent demos included an Android version with a built-in AI agent and a refreshed version of Google Search that uses agent coding to generate a website or app on the fly in response to a search query.
AI coding has wowed the AI world in recent months, even raising hopes that models will one day be able to rewrite their own code in a self-improvement loop. Hassabis says it’s possible, but he doubts it will immediately lead to superhuman-level artificial intelligence.
Progress in other areas of science may require artificial intelligence models to have a deeper understanding of the physical world and even the ability to conduct experiments in it, he says.
Even in the seemingly solved world of coding, Hassabis says it’s notable that artificial intelligence has yet to create a hit app or video game without human support. “I think something is missing,” he says.