Key insights:
Demis Hassabis sits down with Sequoia Capital at AI Ascent 2026 and drops a bold prediction. He believes we are three quarters of the way to Artificial General Intelligence, and that AGI will arrive by 2030.
That timeline is not a guess. It is the same forecast he made back in 2010 when he co-founded DeepMind. The field is, in his words, exactly on track.
So what does that mean for you, your career, and the way you build software? Let's break down the conversation and explore what it signals for the next decade of work.
Demis points to three converging ingredients. Deep learning matured. Reinforcement learning scaled. Compute, in the form of GPUs and TPUs, caught up to the algorithms.
When DeepMind started in 2009, almost nobody in academia believed AGI was possible. Researchers literally rolled their eyes. Today, the same researchers are racing to publish on the topic.
The 20 year mission Demis set in 2010 is now five years from completion. If he is right, the world after 2030 looks nothing like the one we live in today.
It means most of the foundational research is done. The remaining work involves scaling, refining reasoning, improving memory, and pushing toward true agency.
Today's models already write code, analyse documents, and run agentic workflows. The next leap involves systems that plan, reflect, and act over long horizons without supervision.
That leap is what makes the next 24 months critical. The companies and individuals who learn to build with these tools now will shape the era that follows.
You have two choices. Wait and watch, or build and adapt.
The builders winning right now are the ones combining classic automation with AI agents. They use Robotic Process Automation for stable, rule based tasks. They use AI agents for unstructured, judgement heavy work.
If you want a structured way to learn both, the Complete RPA Bootcamp takes you from beginner to pro across RPA, Agentic Automation, Coded Automation, and Computer-Use Agents. You become the person building the AI, not the person it replaces.
Demis has worn many hats. Chess prodigy. Games developer. Neuroscientist. Founder. CEO of Google DeepMind. Nobel laureate.
It looks scattered until you hear him explain it. Every step was a deliberate move toward building AGI.
In the 1990s, games were where the most advanced technology lived. GPUs were designed for graphics engines. AI was the core gameplay component in titles like Theme Park, which Demis built at age 17.
Theme Park sold over 10 million copies. It simulated thousands of visitors, each with an internal economic model. Watching players delight in interacting with that AI convinced Demis to commit his career to the field.
Games funded AI research through the back door. They also taught him a critical startup lesson.
Elixir Studios was Demis' first startup, straight out of college. The team built Republic, a game that simulated an entire country with a million people running on a Pentium PC.
It was too ambitious. The lesson Demis took with him was simple. You want to be five years ahead of your time, not fifty.
That same principle applies to automation careers today. RPA and AI agents are at the five year ahead sweet spot. The tools work, the market is hungry, and the skill gap is wide.
Demis and his co-founders pitched a secret. They had seen something almost nobody else had. Deep learning, just invented by Geoffrey Hinton, combined with reinforcement learning and accelerated compute, could scale.
They felt like keepers of a secret. Even if the bet failed, it would fail in an original way. That conviction attracted high calibre researchers who wanted to work on the most important problem in computing.
The lesson for founders is clear. Conviction backed by genuine insight is what pulls great people in. Hype alone does not.
For Demis, building AGI was never the end goal. The original DeepMind mission statement had two steps. Step one, solve intelligence. Step two, use it to solve everything else.
That second step is where science, medicine, and human flourishing live.
Drug discovery currently averages 10 years per medicine. Demis believes AI will collapse that to months, then weeks, then potentially days.
Isomorphic Labs, the DeepMind spinout, is building on top of AlphaFold to design compounds that bind to specific protein targets without toxic side effects.
The dream is to do 99 percent of exploration in silico. The wet lab becomes only a validation step. If that works, all disease could come into reach, including personalised variations of base medicines.
DeepMind is working on what Demis calls a virtual cell. It is a learned simulator of biology, the same way GraphCast and WeatherNext are learned simulators of weather.
Biology is full of weak signals, correlations, and emergent behaviour. Maths struggles to describe it. Machine learning, Demis argues, is the perfect description language for it.
If you can simulate a cell accurately, you can run thousands of experiments in software before touching a lab bench. That changes everything about how research gets done.
Demis thinks yes. Two new fields are forming.
Imagine running an interest rate change a thousand times across a simulated economy before any central bank acts. That is the kind of decision support tooling Demis sees emerging.
The conversation moves beyond engineering into philosophy. Demis treats these questions with the same seriousness he brings to algorithms.
Einstein showed matter and energy are equivalent. Demis goes further. He believes information has equivalent fundamental status, and may even be primary.
Biology resists entropy by processing information. Structure is organised data. The universe, viewed this way, is an information processing system.
If that view is correct, then AI is not just a useful technology. It is the most direct way to read the language of reality.
Demis refers to himself and his team as Turing's champions. AlphaFold proved that a classical neural network can model a system once thought to require quantum simulation.
Protein folding involves quantum effects at small scales. Yet a classical machine, framed correctly, gets to near optimal solutions. That suggests many problems we assume need quantum computers may not.
Our brains, he suspects, are approximate Turing machines. That makes general intelligence, in principle, achievable on the hardware we already have.
Today, AI is a tool. A telescope for the mind. Demis recommends keeping it that way for as long as possible.
Build the most useful, precise, intelligent tool first. Use it to understand the brain, define consciousness, and tackle the harder philosophical questions.
Agency and consciousness come later. They are the second rubicon, and we should cross the first one with care before approaching the second.
Demis gives a roadmap. AGI by 2030. Drug discovery compressed by orders of magnitude. New sciences emerging. Information as the fundamental substance of reality.
The question is, where do you stand when that wave arrives?
Not the ones who learn to build with it. The ones who treat AI as a tool, integrate it into workflows, and ship real automation will thrive.
The ones who ignore it will watch their roles shrink. This is the same pattern we saw with cloud, mobile, and the internet itself.
Demis is clear that the next year or two are critical. Skill acquisition now compounds for the rest of your career.
Because most enterprise work is still structured, rule based, and audit heavy. RPA handles that reliably and cheaply.
AI agents handle the messy, unstructured, judgement heavy parts. Combined, they form the backbone of modern automation.
You need all four to build production grade automation in 2026 and beyond.
Self study works, but it is slow and scattered. A structured path saves months.
The Complete RPA Bootcamp takes you from beginner to pro across RPA, Agentic Automation, Coded Automation, and Computer-Use Agents. You learn to build the systems Demis is describing, not just consume them.
It is a future proof career switch in the age of AI. Instead of being replaced, you become the one building what replaces outdated workflows.
If Demis is right about 2030, the window to position yourself is now. The conversation with Sequoia partner Konstantine Buhler is worth watching in full. The Sequoia Capital YouTube channel consistently publishes some of the best founder and researcher interviews in the industry, and this one with Demis Hassabis is no exception. Watch the embedded video below to hear it in his own words.