AI Is Turning Biologists Into Solo Software Teams

Derya Unutmaz, an MD and immunologist, is using Codex to build complex scientific applications—flow cytometry analysis, CRISPR design tools, and T-cell signaling simulators—entirely with AI. This isn't pet projects; it's production-grade research tooling that previously required specialized software and engineers.

The striking signal is speed and empowerment. A domain expert with minimal coding skill can now create custom, interactive tools in days. The cost of experimentation collapses, shifting the bottleneck from software development to scientific creativity. For engineering leaders in biotech, this means rethinking how research software gets built.

But hype lurks. Unutmaz's vision of digital twins and curing all diseases in 15 years assumes exponential compute growth that's not guaranteed. The immediate practical value is in accelerating mundane analysis, but full simulation of biology remains speculative. Teams should focus on augmenting existing workflows, not replacing them.

The tradeoff is validation. AI-generated code can produce beautiful outputs, but scientific rigor demands testing against ground truth. The engineering challenge shifts to governance: how to trust, verify, and maintain AI-built tools when the builder may not understand the code. Leaders must balance speed with safety.