Demis Hassabis, the CEO of Google DeepMind, is eager to speak in regards to the coding expertise of his firm’s latest mannequin, Gemini 3.5 Flash. The mannequin has been educated to carry out complicated agentic coding duties: translate giant code bases from one language to a different; discover and repair bugs lurking deep in knotty code; and even write complete working programs from scratch.
Hassabis doesn’t, nonetheless, assume this spells doom for software program builders. “I don’t know why persons are going round speaking with certainty about that,” Hassabis tells WIRED forward of the brand new mannequin reveal at today’s Google’s I/O event.
“Maybe there may be an ulterior motive for placing these messages out; elevating cash or no matter,” Hassabis says. “From my viewpoint, from DeepMind and Google’s viewpoint, if engineers have gotten three or 4 occasions extra productive, then we simply [want to] do three or 4 occasions extra stuff.”
The putting coding skills of the newest fashions has led to widespread worry that AI could also be on the point of eliminating programming roles and different white-collar jobs. Executives at some AI firms have predicted widespread job displacement, whereas some distinguished tech firms, together with Amazon, Salesforce, and Block, have blamed recent layoffs on the usage of AI.
Hassabis thinks that Alphabet, which oversees a number of firms apart from Google, could also be effectively positioned to benefit from a revolution in software program productiveness. “I’ve 1,000,000 concepts, from lab drug discovery to sport design,” he says. “I might like to have some free engineers to go and do these sorts of issues.”
Hassabis says that firms seeking to change builders with AI could also be making a giant mistake. “I believe it is a lack of creativeness—and a lack of information of what is actually going to occur,” he says.
Google revealed a raft of AI stuff at its annual developer occasion. By way of a coding instrument known as Antigravity, Gemini 3.5 Flash presents frontier coding and reasoning capabilities however is quicker and cheaper than the choices of its rivals, Google says. Gemini 3.5 Professional, a extra highly effective new model of its flagship mannequin, will debut subsequent month.
The corporate must catch up with regards to AI coding, which has emerged as a vital and profitable software for the newest AI fashions. Anthropic and OpenAI lead developer adoption with their respective instruments, Claude and Codex, in response to a 2025 Stack Overflow survey.
The corporate additionally demoed an agentic assistant known as Spark that lives in Google’s Cloud and has entry to its apps. The design is supposed to be safer than one thing like OpenClaw as a result of it has restricted entry to private information, Google says.
Different agentic demos included a model of Android with an AI agent inbuilt and a refreshed model of Google Search that makes use of agentic coding to generate a website or app on the fly in response to a search question.
AI coding has captivated the AI world in latest months, even inspiring hope that fashions may someday rewrite their very own code in a self-improvement loop. Hassabis says it’s attainable however doubts that it’ll instantly result in superhuman-level AI.
Progress in different areas of science would possibly require AI fashions to have a deeper understanding of the bodily world and even a capability to carry out experiments inside it, he says.
Even throughout the seemingly solved world of coding, Hassabis says it’s notable that AI has but to supply a blockbuster app or online game with out human assist. “I believe there’s one thing lacking,” he says.
