Demis Hassabis on AGI: What Google DeepMind Thinks Is Coming Next

Sam Altman talks about the singularity like it’s a train that has already left the station. Demis Hassabis talks about it like a scientist who wants to check the tracks first.

Same destination, very different tone. And since Hassabis leads the lab behind Gemini, AlphaFold and a Nobel Prize, it’s worth knowing what he actually says. 2026 has been a big year for him: a Davos clash, a major reshuffle at Google DeepMind and a new institute to argue about AGI in public.

Here’s the decoded version: what he’s said, what has changed, where he differs from OpenAI and what a marketer or founder should do with all of it. (Spoiler: it’s less “buy a bunker” and more “tidy your website.”)

How we wrote this: facts are linked to the reporting we found. We were unable to open some articles in full, so a few details come from search summaries; the PR checklist flags them. Our interpretation is labeled “Our read.” Rumors are labeled as rumors.

The short version: what does Hassabis think about AGI?

Hassabis thinks AGI, meaning AI that matches humans across the full range of cognitive tasks, is probably about the end of this decade, but that today’s systems are not there yet. At Davos in January 2026, Fortune reports he called current systems “nowhere near” human-level AGI and put roughly even odds on it by 2030 (Fortune). In May, Axios headlined that he says we’re “close” (Axios).

Is that a contradiction? Not really. “Close” and “not yet” fit in the same sentence when the finish line is five years away and the track is steep.

What has he said in 2026, month by month?

His message has been consistent in substance, and slightly warmer in tone as the year went on. Here’s the timeline as reported:

A note on the February line: we saw it quoted in search summaries rather than a transcript we could open, so treat the wording as secondhand until you check the original talk.

Why did Hassabis step back from the CEO job?

On August 5, 2026, Hassabis moved from CEO to chair of Google DeepMind and Alphabet’s chief scientist, and Koray Kavukcuoglu took over day-to-day operations. Reporting says the CEO title was dissolved rather than handed over, Kavukcuoglu reports to Sundar Pichai and oversees Gemini model development, and Hassabis keeps leading Isomorphic Labs, the Alphabet drug-discovery company (CNBC; Fortune). Google framed it as the next chapter of its AI momentum (Google).

Two things are fact: the move happened, and Hassabis said he wants to focus on the big picture. One thing is rumor: some reporting ties the reshuffle to a delayed Gemini 3.5 Pro release and pressure to keep pace with OpenAI and Anthropic. Google hasn’t confirmed that link.

Our read: a scientist who believes AGI is imminent is deciding where his time matters most. Whether you read it as vision or a reshuffle under pressure depends on how you feel about Google. Probably both are partly true, and neither changes what your customers do tomorrow.

How does Hassabis’s view differ from OpenAI’s and Anthropic’s?

Hassabis is the careful voice among the three: similar direction, different tempo. At Davos, Fortune reports, Anthropic’s Dario Amodei expected AI to take over most software developers’ work within a year, and Hassabis sounded more cautious. Altman has said the takeoff has already started, and OpenAI has aimed for an automated AI researcher by March 2028 (we covered that in What Sam Altman Is Really Building With ChatGPT; Amodei’s case is in Dario Amodei’s optimistic case for AI).

Two caveats before you pick a winner. First, they define AGI differently, so “we disagree” can sometimes mean “we’re counting different things.” Second, each runs a company that benefits when AI looks both powerful and imminent. Treat all timelines as forecasts from interested parties, including Google’s.

By September the argument had reached the public stage: TechCrunch reports the new DeepMind Institute launched amid a dispute over whether AGI has already arrived, with Nvidia’s Jensen Huang and OpenAI’s Greg Brockman saying it has, while DeepMind co-founder Shane Legg disagrees and keeps a 50% forecast for “minimal AGI” by 2028 (TechCrunch). Even people inside the same lab aren’t aligned on the date.

What is the DeepMind Institute?

The DeepMind Institute is a public platform, launched September 2026, for debating AGI’s impact on society, the economy and safety. Shane Legg, James Manyika and Hassabis are its directors, and it publishes essays whose authors may disagree. The institute states its pieces reflect their authors’ ideas and “should not be read as Google’s official view” (TechCrunch).

Why this is interesting: you don’t launch an institute to debate a risk that you think is five decades away. Our read: the people building this believe the questions are close enough to need a public forum now.

Reality check: what has shipped vs. what is promised?

Promised: AGI around 2030. Shipped: very capable but uneven assistants. Gemini powers Google Search’s AI features, Workspace and a large developer platform, and AI is increasingly useful in science. But the same models still make mistakes people can spot quickly, and Hassabis himself says more breakthroughs are probably needed. Gemini 4 has been referenced in company messaging but, as of this writing, we have not verified a public release, so don’t plan around it.

A simple filter we use with clients:

• Shipped and measurable: AI Overviews and AI Mode in Google, AI assistants that cite sources, agents that complete bounded tasks.
• Announced, not yet proven for most businesses: fully autonomous agents running a department.
• Forecast: AGI itself.

Spend your budget on the first group, run experiments on the second and read the third for entertainment and strategy.

Fear, gain, pain: what this means for your business

• Fear: you wait for “real AGI” before changing anything, while competitors quietly get found in AI answers and automate their busywork.
• Gain: you don’t need AGI to win. Today’s tools already compress research, content, reporting and follow-up from days to hours if the workflow is designed well.
• Pain: every lab gives a different date, the tools change monthly and your team isn’t sure what to adopt first.

The way through is boring and effective: decide on capabilities, not dates.

What this means for your marketing

Three practical implications, all of which hold whether Hassabis is right on 2030 or off by five years.

1. Google remains an AI company first. Its search results increasingly include AI-generated answers, so your content must be easy to quote. Start with the Generative Engine Optimization guide.
2. Agents will do more of the shopping and research. Structured product data, clear pricing and consistent brand facts matter more than clever copy. See Agents Are Shopping for You.
3. Your edge shifts to taste and trust. When everyone has the same models, what differs is your data, your brand voice and your judgment.

What to do this week

1. Pick a capability, not a date: write down the three marketing tasks your team repeats weekly that AI could draft (reports, first-draft content, lead follow-up).
2. Check your AI search presence: ask Gemini, ChatGPT and Perplexity about your category and note who gets cited. The free AI Marketing Stack guide helps you choose tools for each job.
3. Automate one workflow end to end and measure time saved. Don’t wait for a smarter model to start.
4. Set a quarterly reading ritual: one hour to review lab announcements, update assumptions and decide what to test next.
5. Stay skeptical, kindly: when someone gives you an AGI date, ask who benefits and how they define it.

Hassabis may be right about 2030. Altman may be right about sooner. Either way, the customers who ask AI for recommendations are already here.

---

WRITTEN BY

Junaid Mubashar

Founder of an AI-native growth agency. Junaid plans and runs full-funnel marketing (SEO and AI search, paid social, Google Ads) and builds the AI workflows and agents that automate it. Everything on this blog is tested with real campaigns before we recommend it.

About JunaidLinkedInYouTube

Free Toolkit · PDF

🧰 The 2027 AI Marketing Stack

The 37 AI tools and 12 workflow recipes we use to run marketing for clients — organised by funnel stage.

Company website

Get it free →