The Amgen CEO spends his weekends vibe-coding, has required his entire senior executive team to take AI courses, and talks about artificial intelligence with a rare fluency. In a podcast interview with AI pioneer Andrew Ng, he was told: “A lot of times you sound more like an AI person than a biotech CEO.”

The irony is that Bradway runs a company built on the opposite of the autodidact. Amgen’s competitive moat is its experts — scientists who have spent decades learning that molecules are delicate constructions indeed. “The hallways are full of people who have the experience of understanding if you make a substitution of one amino acid for another, there’s a consequence,” Bradway said. “We want to be very careful not to lose that.”

Reconciling those two things — the adaptable generalist and the irreplaceable specialist — is the central challenge of Bradway’s AI moment. And few CEOs are better positioned to navigate it. That’s because Bradway didn’t just start thinking about AI recently. He started in 2012.

That year, Amgen acquired DeCODE Genetics, a small Icelandic company holding longitudinal genomic data on virtually the entire population of Iceland — not a drug or a patent, but a bet that data and AI would eventually transform how medicines get discovered. His peers were skeptical. Even Andrew Ng, who would later become a close collaborator, told Bradway at the time that the genetic data wasn’t “tractable” yet and steered him toward nearer-term AI applications like computer vision. Bradway made both bets and today, an Nvidia SuperPod hums in Reykjavik, Amgen has developed its own protein folding models, and the company is actively pursuing zero-shot antibody design — using AI to engineer drug molecules without prior experimental examples.

The results are tangible, Bradway said. Amgen is now selecting drug candidates for clinical development roughly 50% faster than before. Rather than relying solely on DeepMind’s celebrated AlphaFold, Bradway pushed his team to look past the hype and identify precisely what the model could and couldn’t do for their specific needs — then build on top of it. “We’re not relying on any single approach,” he said.

Amgen told Fortune that it has several “tangible examples” of how AI is additive and supportive to its teams’ work, with a focus on freeing up time for them to focus on what really moves the needle in terms of getting medicines to patients. It has reduced production line clearance time at one manufacturing site from approximately 30 minutes to about two minutes per batch run, and in certain trials, participants were enrolled up to three times more efficiently, according to the company. AI tools and training available companywide include Microsoft Copilot and OpenAI’s ChatGPT Enterprise, alongside investments in training education and responsible-use frameworks. 

His next frontier is agentic AI, and the timeline is explicit: “2026 is the year we all need to figure out what agentic AI is.” The opportunity, as he described it, is profound: scientists today burn hours filling out forms, requisitioning materials, summarizing data — work that agentic systems could absorb entirely, freeing researchers to do what only they can do. “That’s magic,” he said. It’s also where the autodidact-versus-expert tension gets resolved, in Bradway’s telling: AI is liberating, rather than displacing. “Hopefully they have more time to use the essence of their skill,” he said, “and less time devoted to menial things.”

Whether that reassurance lands with employees is another matter. Bradway acknowledged that America broadly doesn’t trust AI and that uncertainty breeds anxiety. “There is a lot of misinformation and there’s a lot of uncertainty and humans hate the unknown, right? Humans don’t like to be kept in the dark and so the more we can share transparently what we’re doing, I think the better.” Still, he admitted, “I don’t have the answer to that, although as you can see, I’m an optimist.”

He’s also, at this point, a proven one. While other CEOs are still debating AI’s return on investment, Bradway is openly baffled by the hand-wringing. “I read a lot and hear a lot about people expressing frustration that they’re not earning a return from the investments that they’re making. I’m a little bit puzzled by that,” he said. Then again, he noted that in biotech, patience isn’t optional and a single molecule can consume a decade-and-a-half, so maybe his “bewilderment” is down to the different calculus. The companies that can’t tolerate uncertainty, he implied, may be precisely the ones that fall behind.

Historically, the process with molecules has relied heavily on iterative laboratory work, manual data review and sequential testing, and AI can help researchers analyze significantly larger datasets, model molecular behavior, predict protein structures, and evaluate candidate molecules earlier in the process. The goal, Amgen said, is to deprioritize less promising candidates earlier. 

“Being late to this party is going to be expensive,” he said. “Being early to the party has the potential for real benefit.”

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.