AlphaFold's breakthrough came not from a smarter algorithm, but from assembling a team that no academic lab was structured to hire. When Google DeepMind's AlphaFold group solved protein structure prediction in 2020, they weren't the most senior team in the field, nor had they worked on the problem the longest. What they had was a deliberate talent strategy that combined machine-learning researchers, structural biologists, chemists, biophysicists, and a large body of research engineers embedded directly into the science.

Why Did 50 Years of Effort Fail to Crack Protein Folding?

The protein folding problem had stalled not because biologists weren't trying hard enough, but because the structure of academic institutions prevented the right people from working together. Academic labs are organized by discipline, and departments hire within disciplinary walls. A structural biologist wanting to hire a software engineer faced resistance from grant bodies, tenure committees, and promotion systems that didn't recognize cross-disciplinary collaboration as legitimate work.

The field had all the raw ingredients for success. Researchers had access to a public dataset of around 200,000 experimentally determined protein structures, decades of theoretical work dating back to Christian Anfinsen's Nobel Prize in 1972, and the CASP (Critical Assessment of Protein Structure Prediction) competition, established in 1994 to provide a shared benchmark. What was missing was a team that could synthesize all of these inputs simultaneously. Labs were full of brilliant scientists writing code themselves, on the side, between experiments. The code became the bottleneck.

How Did DeepMind Build a Team That Academia Couldn't?

DeepMind's competitive advantage in solving a biology problem was not biological expertise. It was the ability to assemble a team that no academic department in the world was structured to hire. The company devoted roughly one third of its total headcount to an internal organization called Research Engineering, led by a senior engineering executive with its own culture and career track.

Half of those engineers built shared tooling like benchmarking infrastructure, data pipelines, and leaderboards that any researcher could use to test a new idea within roughly a day. The other half were embedded inside research pods of three or four scientists, treated as co-creators of the science rather than as service staff. This structure allowed the AlphaFold team to spend the first nine months engineering and curating data, obsessing about accidental data leakage, choosing the right metrics, and building infrastructure that could be used to experiment fast. The deep-learning models came later.

In 2018, after AlphaFold had won the CASP13 competition but had not yet reached the accuracy needed to be useful to working biologists, Demis Hassabis made the talent decision that transformed the project from impressive into transformational. He expanded the team and installed John Jumper as the new research lead, with an explicit goal: redesign the system from scratch with a completely new architecture. That redesign, supported by a multidisciplinary team combining machine-learning experts, great engineers, chemists, biochemists, structural biologists, and biophysicists, became AlphaFold2.

What Made John Jumper the Right Person for the Job?

John Jumper's career path illustrates why academic hiring structures couldn't have assembled the AlphaFold team. Jumper earned a bachelor's degree in physics and mathematics from Vanderbilt University, then received a Marshall Scholarship to Cambridge to pursue a PhD in theoretical condensed matter physics. After a year, he left with a Master's degree because the work wasn't a fit. He then spent three years at D.E. Shaw Research running molecular dynamics simulations on proteins and supercooled liquids. Finally, he earned a PhD in theoretical chemistry from the University of Chicago, working on machine learning for protein folding and dynamics.

By the time DeepMind recruited him in late 2017, Jumper was a physicist who had become a chemist who had become a machine-learning researcher. No academic department in the world is structured to hire that person directly for a tenure-track role. DeepMind's structure was.

How to Assemble a Breakthrough Team Across Disciplines

• Hire for Trajectory, Not Pedigree: Look for people whose careers show evidence of moving across disciplines and learning new fields, rather than those who followed a single, linear path within one department or specialty.
• Create Embedded Engineering Roles: Position engineers and infrastructure specialists as co-creators of science rather than service staff, giving them equal standing and career advancement opportunities alongside researchers.
• Build Shared Infrastructure First: Before diving into the core research problem, invest in benchmarking tools, data pipelines, and leaderboards that allow any team member to test ideas quickly and iterate without bottlenecks.
• Redesign When Needed: Be willing to bring in new leadership and completely rebuild a project's architecture if the initial approach isn't reaching the necessary performance threshold, rather than incrementally optimizing a flawed foundation.

The AlphaFold breakthrough was assessed against 145 other entries in CASP14 and produced a median backbone accuracy of 0.96 ångström. The next-best system was at 2.8 ångström, making AlphaFold roughly three times more accurate and comparable to experimental methods that take a PhD student a year per structure to produce. In October 2024, Hassabis and Jumper shared the Nobel Prize in Chemistry for the work.

"The team had been designed to make those models cheap to run and cheap to throw away," noted the analysis of DeepMind's approach to building AlphaFold.

Source 1

The lesson extends far beyond protein folding. Every CEO now faces the same decision that Hassabis faced: whether to hire and organize teams in ways that fit existing institutional structures, or to redesign those structures to fit the problem. Progress did not stall in biology because biologists weren't trying hard enough. It stalled because the people in the room were the wrong combination for the problem in front of them. A problem becomes structurally unsolvable when the people who could solve it are not allowed to be hired together.