[AI Library] 22 Running a Lab Like a Company
Demis Hassabis, Father of Google's Artificial Intelligence
Part 8. CEO of Google DeepMind
22 Running a Lab Like a Company
Kim Kyung-ran, Kim Kyung-jin
focused on one person's announcement. The post that Google CEO Sundar Pichai published on the company blog was not a simple organizational restructuring notice. It was a declaration that two of the largest intellectual groups in the history of artificial intelligence (AI) were merging into one.
"We are launching a new organization called 'Google DeepMind' by combining Google Brain and DeepMind." This merger fell somewhere between outside critics calling it a 'shotgun wedding' and the internal explanation that it was an 'inevitable evolution.' It came just five months after OpenAI had shaken the world with ChatGPT.
Google was feeling a sense of crisis, and London's DeepMind and California's Google Brain had long been 'rival siblings,' respecting each other while competing. There was only one leader who could unite and lead these two giants: Demis Hassabis. The merger of DeepMind and Google Brain was a meeting of two different 'civilizations.'
DeepMind, founded in London in 2010, was an organization that moved in lockstep under a single, grand mission to 'solve intelligence,' much like an Apollo Project. Google Brain, by contrast, had the atmosphere of a university research lab, valuing academic freedom and openness, where researchers pursued diverse projects following their own curiosity. The task given to Hassabis was immense.
He had to align more than 2,000 of the world's top minds in a single direction. He defined this process as 'reclaiming the speed of a startup without losing the purity of a research lab.' In the early days of the merger, a subtle tension ran between the two organizations.
DeepMind's researchers worried that Google's commercial demands would compromise the essence of their research, while Google Brain's researchers bristled at DeepMind's centralized style of control.
But Hassabis converted this tension into 'productive friction.' He made his case to the researchers: "The resources we have are the best in the world.
When Google's computing power (TPUs) combines with your brains, we can reach AGI faster than any other organization." The Gemini project was, in fact, the first test of this fusion. Born from the combination of DeepMind's reinforcement learning expertise and Google Brain's Transformer architecture technology, Gemini proved what kind of synergy the two organizations could produce once united.
The Engine Room role. In the past, DeepMind was sometimes classified as one of Alphabet's 'Other Bets,' Google's parent company. In other words, it was an investment in the future rather than a revenue-generating unit. But after 2023, DeepMind's standing changed completely.
Hassabis now describes DeepMind as the heart of the massive ship that is Google, the 'Engine Room.' "We supply intelligence to every Google product." This shift brought deep soul-searching about the lab's identity.
Where once the top priority had been writing papers and producing academic results, there was now added pressure to build practical AI models for Search, YouTube, and Android, products used by billions of people worldwide. Clashes between the Product division's demands, Platform stability, and DeepMind's own Research goals were unavoidable. Hassabis used a 'dual-track' strategy to manage these collisions.
On one hand, he focused company-wide resources on developing large language models (LLMs) like Gemini that were directly tied to commercialization. On the other, he fiercely protected a separate space for fundamental scientific research like AlphaFold. He calmed the unease among researchers by emphasizing 'a virtuous cycle in which commercial success funds basic science, and discoveries in basic science dramatically improve the performance of commercial models.'
The relationship with Sundar Pichai. At the center of this massive transformation is the close partnership between Hassabis and Sundar Pichai. Since the merger, Hassabis travels between London and California and speaks with Pichai almost every day. Where the relationship once involved quarterly performance reports, it has become one of a frontline field commander and a supreme commander discussing the state of battle in real time. Pichai delegated full technical authority to Hassabis.
Even within Google, this was an extraordinary level of trust. The 'research autonomy' that Google founders Larry Page and Sergey Brin had promised when they acquired DeepMind was, in this moment of crisis, elevated to a 'strategic partnership.' Hassabis does not hold back from giving Pichai candid counsel not just on technical matters, but on the ethical direction and pace that AI should take.
"We must not rush this technology out the door. But we cannot be late, either." To solve this paradoxical proposition, the two men stay up late every night devising ways to increase speed without compromising on safety.
If Pichai is the shield defending the politics and management of Google's vast empire, then Hassabis is the craftsman sharpening the deadliest spear within it. Google DeepMind, under Hassabis's leadership, is both the largest laboratory in human history controlling and advancing the most powerful technology, and a fiercely contested business battlefield. Hassabis leads 2,000 geniuses through the fog of this uncertain front.
"The Second Shift at Night." As London's night deepens and the city's noise fades, Demis Hassabis's 'real' day begins. During the daytime, he chairs meetings as a CEO overseeing thousands of employees, signs off on paperwork, and untangles complicated organizational problems. But after dinner with his family and putting the children to bed, around 1 a.m. he becomes a 'researcher' again. The 3 a.m. 'Second Shift': the time zone of creativity. Hassabis calls this routine his 'Second Shift.' Running from roughly 1 a.m. to 4 a.m., this window is sacrosanct. He told Fortune magazine:
"I come alive at about 1 a.m. It's hard to be creative during the day. You're constantly being interrupted." During these hours, he does not check email or issue work directives.
Instead, he wrestles with the hardest problems. Breaking through a stuck point in AlphaGo's algorithm, devising a new approach to protein structure prediction, or philosophically contemplating the definition of AGI, these are the main activities of this time. This habit of immersing himself solely in thought amid silence echoes the image of the child prodigy who sat motionless before a chessboard for hours at age four. Unlike many CEOs who practice what is called a 'miracle morning,' rising at dawn to start their day early, Hassabis chose the night.
This shows that he still holds tightly to his identity as an active researcher. Through these late-night hours, he reassembles the scattered fragments of thought from the day and re-sharpens the scientific intuition that was buried under the noise of management. Six hours of sleep is enough for him. Or at least, he tries to believe it is. This night-owl routine of sleeping at 4 a.m. and waking around 10 a.m. has been the hidden engine powering DeepMind's breakthroughs for more than a decade.
Over 2,000 research papers, an h-index of 83, and more than 150,000 citations. The output of this 'second shift' is proven by astonishing academic numbers. Typically, once someone becomes the CEO of a major tech company, they step away from writing research papers. Catching both rabbits, management and research, seems physically impossible.
Hassabis is the exception. As of late 2024, more than 2,000 research papers bear the name Demis Hassabis (including DeepMind's collective output; his personal authorship spans roughly 150 papers). His h-index, a measure of research impact, stands at 83, and his papers have been cited more than 150,000 times by scientists around the world. Even among Nobel laureates, these figures rank near the very top.
The AlphaFold paper published in Nature in 2021 is recorded as one of the most cited papers in the history of science.
What the numbers say is clear. He is not simply an executive who hires scientists and manages them. He personally sets the direction of research, proposes the core ideas, and polishes every sentence of a paper as a lead author. His academic background does not stop at computer science.
A computer science degree from the University of Cambridge, a PhD in cognitive neuroscience from University College London (UCL), and postdoctoral research at MIT and Harvard led him to approach AI not as a mere coding problem but as 'the science of exploring the nature of intelligence.' He is a neuroscientist who uncovered the relationship between memory and imagination, and a computer scientist who opened new horizons in reinforcement learning. His h-index of 83 is proof of a commitment to the pursuit of scientific truth that he never let go of, even amid fierce corporate competition.
Behind the glamorous spotlight of the executive, layers upon layers of solitary hours accumulate, a scientist wrestling with papers under the glow of a desk lamp at 3 a.m. His mission to 'solve intelligence' is not a slogan. It is the product of grueling labor practiced every single night. Google's London office
Kim Kyung-jin
Attorney · Former Member of the National Assembly · AI Policy Researcher
© 2026 Kim Kyung-jin. All rights reserved.