At Google’s I/O event today, Demis Hassabis chose the worst possible moment for lazy AI layoff logic: just as Google DeepMind unveiled stronger coding models, its CEO argued that using AI to dump engineers is a strategic failure, not a technological inevitability.
That is the right argument. Companies treating AI productivity as a permission slip to cut payroll are thinking too small. Hassabis told Wired that if engineers become “three or four times more productive,” Google and DeepMind would rather “do three or four times more stuff.” That is not sentimental labor politics. It is a product strategy.
“I think it's a lack of imagination—and a lack of understanding of what's really going to happen,” Hassabis told WIRED.
The timing matters because Gemini 3.5 Flash is being pitched as exactly the kind of model that makes executives reach for headcount spreadsheets. Google says it can handle complex agentic coding tasks: translating large code bases, finding deep bugs, and even writing operating systems from scratch. If that is the tool, the choice becomes stark. Use it to build more, or use it to shrink ambition.
At I/O Today, Hassabis Turned Gemini’s Coding Pitch Into a Management Test
Google did not show up to I/O with a modest autocomplete story. It introduced Antigravity, a coding tool through which Gemini 3.5 Flash offers what Google describes as frontier coding and reasoning capabilities, while being faster and cheaper than rival offerings. A more powerful Gemini 3.5 Pro is expected next month.
That makes Hassabis’ comments sharper. He is not denying that AI coding tools are getting stronger. He is saying the obvious next step should not be replacing the people most able to use them.
His argument lands because the source material names a real tension. WIRED reports that Amazon, Salesforce, and Block have blamed recent layoffs on AI use. Hassabis’ response is basically: why would you discard the people who can turn productivity gains into new products?
This is especially relevant because Google is trying to gain ground in AI coding. WIRED cites a 2025 Stack Overflow survey showing Anthropic and OpenAI lead developer adoption with Claude and Codex. That is the competitive frame behind our earlier context piece, Google I/O Puts Gemini on Trial as Claude Grabs Devs. Google needs developers to believe Gemini is useful. Firing the class of people who can validate and extend that usefulness would be a strange victory lap.
The First Ripple Effect: Productivity Gains Should Become Product Gains
The strongest version of Hassabis’ case is not “save every job.” It is “do not confuse efficiency with strategy.”
If Gemini 3.5 Flash makes engineers faster at translating code, debugging large systems, or building software from scratch, the obvious growth-minded response is to assign those engineers to more ambitious work. Hassabis gave examples: “lab drug discovery” and “game design.” Those are not random hobbies. They point to a broader theory of AI inside Alphabet: more capable tools should widen the menu of possible projects.
That logic also explains why the layoff-first approach is so thin. Cutting engineers may improve a cost line. But it can also cap the number of experiments a company can run with its new tools. AI does not automatically decide which products are worth building, which bugs matter most, which architecture trade-offs are acceptable, or which weird edge case will become tomorrow’s crisis.
Analysis: In coding-heavy organizations, AI productivity is only valuable when converted into shipped work, tested systems, and better decisions. A model that can write code faster still needs people to define the goal, review the output, catch failures, and decide when “working” is not good enough. Hassabis is saying the scarce asset is not raw code generation. It is judgment plus ambition.
After the Demos, the Human Bottleneck Still Looks Real
Google also demoed Spark, an agentic assistant that lives in Google Cloud and has access to its apps. WIRED says Google designed it to be safer than something like OpenClaw because it has limited access to personal data. Other demos included an Android version with an AI agent built in and a refreshed Google Search that can use agentic coding to generate a site or app in response to a search query.
That is a lot of automation. It is also a lot of surface area for mistakes.
The more agentic these systems become, the more companies need experienced people supervising them. Not because humans are magical. Because software organizations run on context: old architectural choices, customer constraints, security assumptions, product priorities, and hard-won scar tissue from past failures.
Analysis: Mass AI layoffs risk removing precisely the people who know where the bodies are buried in the codebase. A model may find a bug. A senior engineer may know why the tempting fix breaks something more important. A model may translate a codebase. A team still has to decide whether the translation preserves performance, security, and maintainability.
That is why the “AI replaced the role” framing is too crude. Some tasks may disappear. But the work of directing, evaluating, and improving AI output becomes more important as the tools become more capable. For related context on Google’s cheaper agent push, see Cheap AI Agents: Google’s Gemini 3.5 Flash Bets Big.
The Next Management Decision Is Retraining Before Downsizing
If Hassabis is right, the smart corporate response is not to freeze in place. It is to move workers into higher-value workflows before making cuts.
That means training engineers to work with agentic coding tools, redesigning review processes around AI-generated code, and giving teams incentives to find new projects rather than merely compress old ones. It also means measuring output differently. If a team becomes three times faster, leadership should ask what new backlog can now be attacked, not only how many seats can be removed.
This is where executives often reveal their real priorities. A company that sees AI as a growth engine will redeploy talent into new product lines, research work, and faster iteration. A company that sees AI as a cost-cutting slogan will ask finance to harvest the savings before product teams can compound them.
Hassabis’ own phrasing matters here. “I'd love to have some free engineers to go and do those kinds of things,” he told WIRED. Free engineers, not fewer engineers.
The Counterargument Is Real: Some Roles Will Change or Vanish
The best counterargument is that companies cannot pretend automation has no labor consequences. If models can perform more coding tasks, some work will be reduced. Businesses face pressure to spend carefully, and ignoring efficiency gains would be unserious.
But that does not justify reflexive AI layoffs.
There is a difference between eliminating obsolete work after a serious transition plan and using AI as a blanket explanation for downsizing. The first requires specifics: which tasks changed, which teams were retrained, which roles were redeployed, and why cuts were unavoidable. The second turns AI into corporate fog.
Hassabis does not claim AI coding has solved everything, either. WIRED reports that he doubts self-improving code loops will immediately produce superhuman-level AI, and he says progress in science may require models to understand the physical world more deeply and perform experiments within it. Even in coding, he points out that AI has not yet produced a blockbuster app or video game without human help.
“I think there's something missing,” Hassabis said.
That missing piece is exactly why treating workers as instantly replaceable is reckless.
The Next I/O-Scale Test Is Whether AI Expands Ambition
AI layoffs reveal a leadership choice. The technology is powerful, but the distribution of its gains is not automatic. Executives decide whether productivity becomes more research, more products, more experiments — or just fewer employees doing more under the same deadlines.
Google’s own I/O slate makes the choice concrete. Gemini 3.5 Flash, Antigravity, Spark, agentic Android, and agentic Search all point toward software that can act with more autonomy. Our coverage of AI Agents Grab Google Search—and Start Watching You sits in that same debate: as agents gain reach, supervision and accountability matter more, not less.
CEOs and boards should stop measuring AI success only by headcount reduction. Measure it by new capabilities. Measure it by better products. Measure it by whether teams can pursue ideas that were previously too expensive, too slow, or too technically painful.
If AI is truly transformative, the goal should be bigger than cheaper labor. The companies that win will not be the ones that merely cut fastest. They will be the ones that use AI to enlarge what their people can build.
The Bottom Line
- Hassabis frames AI coding tools as a way to increase output, not replace engineers.
- Google’s Gemini coding push makes the management choice around AI productivity more urgent.
- The debate highlights a growing split between companies using AI for growth and those using it to cut costs.
