Demis Hassabis built an AI to beat his five-year-old brother at Othello in 1988, then spent 35 years turning that childhood trick into the infrastructure running Google's bet on the agent economy.
The Summary
- DeepMind's CEO reflects on the tenth anniversary of AlphaGo, the AI that mastered Go and signaled that machines could handle complexity beyond brute-force computation
- That childhood Othello program wasn't just coding practice. It was the conceptual seed for autonomous agents that "go off and do something on your behalf"
- Google merged DeepMind with Google Brain in 2023, putting Hassabis in charge of Gemini, the model now embedded in products serving billions of users
The Signal
The AlphaGo moment in 2016 wasn't just about a board game. It was proof that AI could develop intuition, strategy, and novel approaches to problems that couldn't be solved by checking every possible move. Go has more possible board positions than atoms in the observable universe. You can't brute-force your way to victory. You need something closer to judgment.
That's the bridge between games and agents. Hassabis frames games as "a microcosm of something important in real life", a safe sandbox where AI can practice making consequential decisions without real-world stakes. The muscle memory built in these controlled environments transfers to messier problems: drug discovery, protein folding, energy grid optimization.
"It's incredible that you can make a program that's inanimate and it can go off and do something on your behalf."
Fast forward from that 1988 Othello program to 2025. Hassabis now runs Google DeepMind, the merged entity combining DeepMind's research chops with Google Brain's production infrastructure. Gemini, the model under his watch, isn't playing games anymore. It's scheduling meetings, writing code, analyzing spreadsheets, and increasingly making decisions without waiting for human input. The childhood epiphany scales.
The timing matters. Google folded Brain into DeepMind in 2023, the same year ChatGPT kicked off the agent race. By putting Hassabis in charge of both research and deployment, Google signaled it was done treating AI as a lab curiosity. The company that missed the social media boat and fumbled cloud early isn't making the same mistake with agents. Gemini is now the default intelligence layer across Gmail, Docs, Search, and Android. Billions of people interact with Hassabis's work daily, whether they know his name or not.
Key developments since AlphaGo:
- DeepMind solved protein folding with AlphaFold, compressing decades of biological research into months
- The merger gave Hassabis control over Google's entire AI stack, research through production
- Gemini went from research project to product infrastructure in under two years
The throughline from Othello to Gemini is agency. Not the philosophical kind. The practical kind. Software that acts on your behalf while you do something else. That's the kernel of Web4. Your agent handles the grunt work. You handle the judgment calls. Or eventually, the agent handles more of those too.
The Implication
Watch what Google DeepMind ships next, not what it publishes. The research-to-product cycle has compressed. AlphaGo took years to move from breakthrough to application. Gemini updates ship in weeks. Hassabis spent a decade proving AI could master complexity. Now he's spending the next decade embedding that complexity into tools that run businesses, manage workflows, and make decisions at scale.
If you're building in the agent space, you're competing with a team that's been training for this since before most founders were born. The game-playing research wasn't a detour. It was the warmup.