If engineers become three times more productive, we do three times more work
At the threshold of a new technological era, Google DeepMind's Demis Hassabis offers a quieter, more expansive vision than the one dominating headlines: that artificial intelligence, rather than rendering human ingenuity obsolete, may simply enlarge the canvas on which it operates. Speaking ahead of Google's I/O conference in May 2026, he challenged the assumption that efficiency must translate into elimination, arguing instead that a more productive engineer is an invitation to dream bigger. His is a philosophy of abundance over subtraction — though the world is still deciding which story to believe.
- Fear of programmer obsolescence has reached a fever pitch, with major companies like Amazon and Salesforce already citing AI as a factor in layoffs.
- Gemini 3.5 Flash can translate entire codebases, hunt hidden bugs, and write operating systems from scratch — capabilities that have genuinely unsettled the industry.
- Hassabis pushes back hard, calling doomsday predictions a failure of imagination and suggesting some forecasters may have financial motives for stoking anxiety.
- Google's answer is a suite of new tools — Antigravity, Spark, an agentic Android assistant — positioning AI as a force multiplier rather than a workforce replacement.
- Even Hassabis concedes a gap remains: no AI system has yet shipped a successful application or game without human hands guiding it, leaving the question of true autonomy unresolved.
Before Google's I/O developer conference in May 2026, Demis Hassabis made a case that has grown uncommon in Silicon Valley: AI will not eliminate programmers. It will make them dramatically more capable — and the right response to that capability is ambition, not downsizing.
The backdrop was the unveiling of Gemini 3.5 Flash, a model that can translate entire codebases between programming languages, surface deeply buried bugs, and write operating systems from scratch. These are not demonstrations of novelty. They represent a genuine shift in what machines can do — and they have fed a wave of anxiety across the industry. Amazon, Salesforce, and Block have each pointed to AI adoption when explaining recent layoffs, and some AI leaders have openly forecast mass displacement.
Hassabis dismissed that framing. If engineers become three or four times more productive, he argued, the answer is to pursue three or four times as much work — drug discovery, game design, the long list of projects Alphabet has never had the engineering bandwidth to attempt. Companies replacing developers with AI, he said, are making a mistake born of limited imagination. They misunderstand what the technology actually is: not a substitute, but a multiplier.
Google's new tools embodied this philosophy. Antigravity, a coding assistant within Gemini 3.5 Flash, is positioned as faster and cheaper than rivals from OpenAI and Anthropic — companies that currently lead in developer adoption. Spark, an agentic cloud assistant, can navigate Google's applications while limiting exposure to personal data. Google also showed an Android agent and a reimagined Search capable of generating websites and apps on the fly.
Still, Hassabis acknowledged a meaningful ceiling. Despite months of speculation about AI systems rewriting their own code in self-improving loops, no model has yet delivered a working application or game without human involvement. Something, he said, is still missing — perhaps a deeper understanding of the physical world, perhaps something harder to name. The technology is powerful. It is not, at least not yet, a replacement for human judgment.
Demis Hassabis, the head of Google DeepMind, sat down with WIRED before his company's latest product reveal and made a case that has become increasingly rare in Silicon Valley: artificial intelligence will not destroy programming jobs. Instead, he argued, it will make developers dramatically more productive—and companies should use that windfall to build more ambitious things, not to shrink their engineering teams.
The occasion was Google's I/O developer conference in May 2026, where the company unveiled Gemini 3.5 Flash, a new AI model trained to handle sophisticated coding tasks that would have seemed like science fiction a few years ago. The system can translate entire codebases from one programming language to another. It can hunt through thousands of lines of tangled code and find the bugs hiding deep inside. It can even write operating systems from scratch. These are not parlor tricks—they represent genuine technical capability at a scale that has spooked the industry.
That spookiness has fueled a wave of anxiety about programmer obsolescence. Some AI company leaders have openly predicted mass job displacement. Amazon, Salesforce, and Block have attributed recent layoffs partly to AI adoption. The fear is understandable: if machines can now do the work of a senior engineer, why keep the senior engineer around?
Hassabis rejected this logic outright. "I have no idea why people go around talking with such certainty about that," he told WIRED. He suggested that doomsayers might have ulterior motives—raising money, perhaps, or generating headlines. His own view, and Google's, was simpler: if engineers become three or four times more productive, the answer is not to fire them. It is to do three or four times as much work. He pointed to his own wish list: drug discovery in laboratories, video game design, and countless other projects that Alphabet could pursue if it had more engineering capacity. "I would love to have engineers freed up to go do that kind of thing," he said.
Companies that try to replace developers with AI, Hassabis argued, are making a fundamental mistake rooted in a lack of imagination. They do not understand what is actually happening. The technology is not a substitute for human programmers. It is a force multiplier.
Google's new tools reflected this philosophy. Antigravity, a coding assistant built into Gemini 3.5 Flash, offers coding and reasoning capabilities that Google says are faster and cheaper than competing offerings from OpenAI and Anthropic—companies that have seized the lead in developer adoption, according to a 2025 Stack Overflow survey. Google also demonstrated Spark, an agentic assistant that lives in Google's cloud and can access the company's applications with limited exposure to personal data. The company showed off a version of Android with an embedded AI agent and a redesigned Google Search that uses agentic coding to generate websites and applications on the fly in response to search queries.
Yet even Hassabis acknowledged limits to what AI coding can do right now. He noted that despite months of hype about AI models rewriting their own code in self-improving loops, no AI system has yet produced a successful application or video game without human involvement. "I think something is missing," he said. Progress in other scientific domains might require AI models to develop a deeper understanding of the physical world and the ability to run experiments in it. The technology is powerful. It is not, he suggested, a replacement for human judgment and creativity. Not yet, anyway.
Citas Notables
I have no idea why people go around talking with such certainty about that— Demis Hassabis, on predictions of programmer job elimination
Companies that try to replace developers with AI are making a fundamental mistake rooted in a lack of imagination— Demis Hassabis, on misuse of AI coding tools
La Conversación del Hearth Otra perspectiva de la historia
When Hassabis says engineers will become three or four times more productive, what does that actually mean in practice? Is he talking about writing code faster, or something else?
It's broader than just typing speed. Gemini 3.5 Flash can handle entire refactoring projects—moving code from one language to another, finding buried bugs, building systems from scratch. A developer might spend a week on something that now takes a day. But the real shift is that the engineer's attention moves upstream: instead of grinding through routine work, they think about what to build next.
That's a generous reading. Couldn't a company just say, "We need the same output in half the time," and cut the team in half?
Absolutely. And some companies are doing exactly that. Hassabis is arguing they're being shortsighted—that the smarter move is to keep the team and tackle harder problems. But that requires a company to have vision and ambition beyond the next quarterly earnings report.
He mentioned that no AI has yet built a successful app or game without human help. Why does that matter?
Because it suggests the bottleneck isn't code-writing anymore. It's everything else—deciding what to build, understanding users, making a thousand small judgment calls that require taste and intuition. The AI is a tool that handles the mechanical parts. The human is still the one steering.
So his argument is that AI amplifies human capability rather than replacing it?
That's the thesis. But it only works if companies actually believe it and act on it. The fear in the industry is that most won't. They'll see productivity gains and treat them as an opportunity to cut costs, not to dream bigger.
And Google? Is Hassabis confident his own company will make the right choice?
He seems to be. He's got a long list of things he'd love to do if he had more engineering capacity. Whether Alphabet actually frees up resources to pursue those ambitions is a different question. Announcements and actions don't always align.