Demis Hassabis — Nobel laureate, DeepMind CEO, and the most measured voice in the AI industry — just dropped the S-word at Google I/O 2026.
“When we look back at this time,” Hassabis told the audience, “I think we will realize that we were standing in the foothills of the singularity.”
He also said AGI is “just a few years away.”
This is the same man who, when we last checked in on his predictions, was talking about world models and continual learning as “the pieces starting to fit together.” That was April. Two months later, he’s using singularity language. The acceleration isn’t just in the models — it’s in the rhetoric.
What else happened at I/O 2026
Hassabis’s comments didn’t happen in a vacuum. They anchored a product keynote that included:
-
Gemini Omni — A new “world model” that understands gravity, fluid dynamics, and kinetic energy. It generates and edits photorealistic video from natural language prompts. First version (Gemini Omni Flash) rolling out to AI Plus, Pro, and Ultra subscribers now, also integrating into YouTube Shorts. All outputs carry mandatory SynthID watermarks.
-
Gemini 3.5 Flash — Four times faster than frontier competitors, outperforming the previous 3.1 Pro generation. Replaces older models as the default in the Gemini app globally.
-
Antigravity 2.0 — An “agent-first” platform that uses Gemini 3.5 Flash to build fully functional applications and operating systems from conversational prompts. This is Google taking direct aim at OpenAI’s Codex and Anthropic’s Claude Code — automating the entire development workflow, not just assisting with code snippets.
-
Gemini Spark — A personal AI agent that runs 24/7 in the background, executing multi-step digital tasks even when your laptop is closed.
-
Search overhaul — The biggest structural update to Google Search in 25 years, replacing blue links with conversational AI, information agents, and generative UI.
When someone says “singularity” while unveiling a product suite that literally replaces the web’s front door, you’re allowed to question the timing.
The credibility problem
Hassabis has earned credibility. He’s a Nobel laureate. He built AlphaFold. DeepMind’s track record on scientific AI is genuinely impressive. When he says world models are progressing fast, that’s worth taking seriously.
But “foothills of the singularity” is not a scientific statement. It’s a marketing statement. It’s the kind of language that primes regulators to take AI seriously (good) while also priming consumers to buy Google’s latest products (convenient).
The same week Hassabis invoked the singularity, a BBC investigation showed that a journalist could manipulate Google’s AI into declaring him a world-champion hot-dog eater in 20 minutes. The “foothills of the singularity” apparently still have a spam problem.
What’s actually new
Gemini Omni is legitimately interesting. A world model that simulates physical reality — not just text, not just images, but physics — is a meaningful step beyond large language models. Hassabis has been talking about world models since April, and this is the first concrete product built on that vision.
The question is whether world models that understand gravity and fluid dynamics actually lead to AGI, or whether they lead to better video generation and more convincing product demos. The answer, honestly, could be both — but we shouldn’t conflate them.
Antigravity 2.0 is also notable. Building full applications from prompts is the logical endpoint of the coding agent trend, and Google’s scale gives it distribution advantages that OpenAI and Anthropic can’t match. If your Gmail, your calendar, your documents, and your search engine all run Gemini, the coding agent that’s integrated with all of them has a structural edge.
Why it matters
Hassabis’s shift from “the pieces are starting to fit together” (April) to “foothills of the singularity” (May) tracks with Google’s escalating product ambitions. The more Google positions itself as an AGI company, the more leverage it has in regulatory conversations (only serious companies should build AGI), in talent wars (come build the singularity), and in consumer markets (buy our products, we’re literally at the foothills).
This isn’t necessarily wrong. Progress in AI is real. World models matter. Agent capabilities are expanding. But the rhetoric and the product roadmap are now so tightly coupled that it’s hard to tell where the science ends and the sales pitch begins.
For New Zealand, the relevance is direct: Google’s AI products — from Search overhauls to Gemini agents — will shape how millions of Kiwis access information. When the company running that infrastructure says it’s at the “foothills of the singularity,” that’s not just a philosophical claim. It’s a statement about how much control over information they intend to have.
The foothills, for the record, are where the guides start charging extra.
Related: Hassabis: 2026 is the breakthrough year · Google Search’s biggest AI overhaul in 25 years · BBC: Google’s AI can be manipulated in 20 minutes
