The Hassabis Map: How to Plan for Twenty Years Without a Calendar

TL;DR: Demis Hassabis didn't predict the future. He built a map of problems that mattered but weren't yet solvable, then waited for the technology to catch up. The protein folding breakthrough took four years of work—and twenty years of preparation. The lesson: plan by trigger conditions, not timelines. Hunt for root nodes, not leaves. And never confuse following trends with following opportunity.

James here, CEO of Mercury Technology Solutions. Hong Kong — July 2026

In 2016, AlphaGo beat Lee Sedol, and the world learned Demis Hassabis's name. What almost no one noticed: on the flight back from Seoul to London, he launched the protein folding project that would win him the 2024 Nobel Prize in Chemistry.

Four years later, AlphaFold2 solved a fifty-year-old problem. But the real story isn't those four years. It's the twenty years before them.

Hassabis first encountered the protein folding problem as an undergraduate at Cambridge in the 1990s. He was studying computer science, not chemistry. He wouldn't start DeepMind for another decade. He wouldn't touch protein folding for two.

Most people look at this and see a career with weird jumps: games, neuroscience, AI, biology. But pull the camera back, and every move was deliberate. The computer science degree, the cognitive neuroscience PhD, the game company, the AI lab—each piece was a move on a board only he could see.

If Hassabis had studied chemistry as an undergraduate, he probably wouldn't have solved protein folding. That's the counterintuitive part. The breakthrough required someone who could hold multiple fields in their head simultaneously, who understood what AI could do before AI could actually do it.

This isn't a biography. It's a deconstruction of how he thinks about planning—and what the rest of us can steal from it.

The Crack in the Ice

Hassabis operates in what I call the crack—the gap between "impossible" and "obvious."

On one side, you have problems everyone is working on. Hot topics. Trending fields. The stuff grant committees fund and conference panels debate. These are crowded. The returns are competed away before you start.

On the other side, you have problems everyone agrees are important but currently unsolvable. The frontier is too far. The tools don't exist. These are graveyards of abandoned PhD theses and failed startups.

The crack is the narrow band in between: important problems that are almost solvable. Not yet. But soon. The technology trajectory is clear, even if the timeline isn't.

Hassabis didn't predict that protein folding would be solvable in 2016. He couldn't have. What he did was maintain a map of problems in that crack and track which ones were approaching solvability as AI capabilities advanced.

This requires a specific kind of attention. Not focus—sensitivity. You're not working on these problems yet. You're watching them. Feeling for temperature changes in the technological environment that might thaw one of them.

The more you track the crack, the better your intuition becomes. You start to distinguish between real capability shifts and hype cycles. You learn to smell when a field is about to tip from "impossible" to "expensive" to "routine."

Most people don't even know the crack exists. They're either chasing what's hot or avoiding what's hard. Hassabis built his entire career in the space between.

Root Nodes, Not Leaves

There's a second filter, and it's where most smart people fail.

Hassabis doesn't solve just any problem in the crack. He hunts for root nodes—problems that sit at the foundation of entire trees of downstream applications.

Think of scientific problems as a massive tree. The leaves are the visible, immediate questions: How do we improve this specific drug? How do we optimize this particular material? These are important, but they're narrow. Solving one leaf helps one branch.

The roots are different. They're the fundamental problems whose solution unlocks entire forests. Protein folding isn't just about proteins. It's about understanding the machinery of life itself. Solve it, and you don't just get better drugs—you get a new way of doing biology.

Most scientists work on leaves because leaves are safer. The path is clearer. The publications come faster. The grants are easier to justify. Root nodes are terrifying because they're hard, and because failure means years of work with nothing to show.

But here's the thing about root nodes: they're only approachable from the crack. By the time a root node becomes obviously solvable, everyone is already working on it. The advantage comes from seeing it before it's obvious—from having the map and the sensitivity to know when the moment arrives.

Hassabis didn't pick protein folding because it was trendy. He picked it because it was a root node he had been tracking for twenty years, and in 2016, the crack finally opened wide enough.

The Bayesian Strategy

Here's where it gets tactical.

Hassabis doesn't plan by calendar. He plans by evidence. Every project—successful or failed—updates his model of what AI can and cannot do.

AlphaGo wasn't just a publicity stunt or a research milestone. It was a probe. It tested whether deep reinforcement learning could master a problem of staggering complexity. The success of AlphaGo provided evidence that AI had crossed a threshold. That evidence updated Hassabis's map. It told him: the tools are ready for protein folding now.

This is Bayesian planning. You hold hypotheses about technological capability. Every project generates evidence that confirms or challenges those hypotheses. The hypotheses update. The map redraws itself. The next move becomes obvious only in retrospect.

This is why Hassabis didn't launch the protein folding project in 2010 or 2012. The evidence wasn't there yet. The tools weren't ready. Starting earlier wouldn't have made him faster—it would have made him fail.

Compare this to how most organizations plan: "Year 1: build game AI. Year 3: beat world champion at Go. Year 6: solve protein folding." This is calendar planning, and it's garbage. Breakthroughs don't arrive on schedule. They arrive when the underlying conditions make them possible.

Planning by timeline is planning for fantasy. Planning by trigger condition is planning for reality.

The Two Ways Planning Fails

I've watched this pattern destroy careers and companies. Two failure modes, both common, both fatal.

Failure Mode 1: Chasing the Heat

In 2016, after AlphaGo, I watched a philosophy professor pivot to AI. Created a WeChat group called "AI Discussion Group." Eighty percent of the members had no AI background. I got invited because I had written a popular (and wrong) prediction about the Lee Sedol match.

The professor gave talks. Applied for grants. Spent two years on "AI and philosophy." Produced nothing. When ChatGPT hit in 2022, he pivoted back to the same topic. Still nothing.

This is heat-chasing. You see something trending, you run toward it. But by the time you arrive, the crowd has already picked it clean. You're not early—you're late, and you're under-equipped.

Failure Mode 2: Locking the Timeline

The opposite error is rigid long-term planning. You set milestones: Q1 deliverable, Q2 review, Q3 integration. This works for engineering projects with known parameters. It fails catastrophically for research, innovation, or any domain where the path is foggy.

Why? Because real breakthroughs don't respect your Gantt chart. If you force results to match timelines, you get one of two things: fabricated "milestones" that are just packaging, or trivial work that was predictable from day one. Neither is breakthrough. Neither matters.

Hassabis avoids both traps. He doesn't chase heat—he's already positioned where the heat will be, before anyone feels it. And he doesn't lock timelines—he locks trigger conditions. "When X capability is demonstrated, then Y project launches."

The Twenty-Year Move

Let's talk about the actual timeline, because this is where most people misunderstand Hassabis entirely.

1990s: Undergraduate at Cambridge. Learns about protein folding. Recognizes it as important. Stores it on his mental map. Does nothing about it.

2000s: Builds game AI company. Sells it. Gets PhD in cognitive neuroscience. Not obviously related to protein folding. But he's building tools and understanding.

2010: Co-founds DeepMind. Focuses on general AI. Still not protein folding.

2016: AlphaGo beats Lee Sedol. On the flight home, launches protein folding project. Four years later, AlphaFold2. Four years after that, Nobel Prize.

The gap between encountering the problem and solving it was twenty-five years. But the active work was only four. The other twenty-one were preparation, positioning, and waiting for the crack to open.

This is what I mean by planning without a calendar. Hassabis didn't schedule "solve protein folding" for 2020. He maintained a living map, updated it continuously with evidence from every project, and pulled the trigger when the conditions were right.

The Go match wasn't a prerequisite for protein folding in any technical sense. You don't need to master board games to fold proteins. But the Go match provided the evidence that the tools were ready. It was a trigger condition, not a stepping stone.

What This Means for You

You're probably not going to win a Nobel Prize. Neither am I. But the Hassabis framework applies to any domain where the timeline is uncertain and the stakes are high.

Build a problem map. Not a todo list—a map. What are the root-node problems in your field? Which ones are in the crack? Which ones are approaching solvability? Track them. Update them. Let them sit for years if necessary.

Plan by trigger, not timeline. "When we hit X metric, we expand to Y market." "When this technology demonstrates Z capability, we launch this product." Not: "We launch in Q3."

Update your model with every project. Success teaches you what's possible. Failure teaches you what's not. Both are evidence. Both update your map.

Hunt root nodes. The obvious problems are crowded. The fundamental problems are where leverage lives. But you need the map and the patience to wait for the crack to open.

Ignore the heat. By the time something is trending, the opportunity is already priced in. Your edge is in seeing it before the crowd, not running with the crowd.

The Go Move

There's a moment in the second game of the Lee Sedol match that Go players still talk about. Move 37. Lee Sedol got up, went to the bathroom, thought for fifteen minutes. It was a move no human would play. Its value wasn't visible in the moment—it manifested twenty moves later.

Hassabis's undergraduate encounter with protein folding was his Move 37. It seemed irrelevant to the game he was playing. Its value wouldn't be visible for twenty years.

Most people can't hold a move in their head for twenty minutes, let alone twenty years. They need immediate feedback, visible progress, quarterly validation. But root-node problems don't reward quarterly thinking. They reward the ones who can wait.

The question isn't whether you have Hassabis's IQ. You probably don't. I certainly don't. The question is whether you can build the map, maintain the sensitivity, and wait for the crack to open.

That's a skill. It can be learned. It can be practiced. And in a world obsessed with quarterly results and trending topics, it's the closest thing to a structural advantage you'll find.

Mercury Technology Solutions: Accelerate Digitality.