Derek Wilcox, Chief Information Officer (CIO) of Discovery Group, refuted the notion that artificial intelligence will soon render software developers obsolete. He stated that the productivity gains his teams are observing when using AI-based coding tools are real, but currently represent only a small fraction of the promises made by the loudest voices in the industry.
During an interview for the TechCentral’s Meet the CIO podcast, Wilcox reported that in some areas of Discovery, efficiency has increased by approximately 20–25% thanks to AI coding assistants. However, he immediately qualified this figure, noting that employees at large corporations typically spend only 25–30% of their time directly writing code; the rest is dedicated to meetings, code reviews, testing, and interactions with business users.
Wilcox emphasized that Discovery develops a significant portion of its software internally across its largest divisions and acquires little traditional enterprise software in healthcare and life sciences.
Obstacles to fully autonomous coding
According to Wilcox, there are three problems hindering the achievement of fully agentic coding with AI: explainability, efficiency, and the ability to support AI-generated code. None of these issues have been resolved to date. He noted that anyone who has worked on a large project using AI-generated code will discover after a year, following model drift, that they are left with a product that even previous AI cannot accurately identify as its own.
This phenomenon is more dangerous than it seems because code represents not just a set of executable instructions, but also the systematization of a company's business processes. Wilcox expects that significant human oversight will remain necessary in the foreseeable future.
Although senior developers at Discovery use all major frameworks for agentic coding, the group has invested heavily in constraining the capabilities of these tools. These efforts aim to control agent actions, ensuring that their outputs comply with Discovery's coding standards, security and privacy requirements, and integrate with existing infrastructure.
The importance of 'scaffolding'
Wilcox acknowledged that calculations differ for startups: a company founded today would have to ignore the need to generate almost all of its code using AI. However, in large regulated enterprises where code touching numerous legacy systems and teams must be written, Discovery consciously maintains a conservative approach—a trend he observes among other CIOs.
His broader bet is that the winners will not be the companies that gain access to the best cutting-edge models. The winners will be those who build what is becoming increasingly known as 'scaffolding'—a system that allows for safe and scalable AI implementation. He added that companies investing in this now will be the future leaders in AI.
This view is reflected in Discovery's hiring approach. Wilcox believes the industry has completed the decade where mere coding ability was sufficient to get a job. Coding remains important, but fundamental knowledge provided by a computer science degree is equally vital: understanding architectures, algorithm principles, design-oriented approaches, as well as risk management and compliance. Add to this the ability to work in a team and communicate, and the most crucial remaining competency is continuous learning. He warned that building a career based solely on a short programming course will be harder in the coming decade than in the last.
Discovery's annual graduate hackathon, where teams from universities and training academies compete over the weekend, showed him how advanced the tools have become: teams can now release a full product in two days, which was unattainable eight or ten years ago.
He was more surprised by the upskilling of existing staff. Discovery provided its IT specialists with free access to well-known online academies, and Wilcox expected only 3–4% to take advantage of this. Instead, 50–60% of employees completed at least one AWS or Google program. Attracting a specialist who is an absolute expert in agent security or advanced AI models is nearly impossible today. However, hiring a graduate capable of quickly mastering these skills is quite feasible.
