You Don't Have to Choose Between Leadership and Code

A few months ago, a CTO I know told me he hadn’t written a single line of production code in six years. He said it with a hint of pride, the way executives sometimes signal they have successfully made the transition from maker to manager. I understood the logic. His time had moved to strategy, budgets, stakeholder management and hiring. Code was for the team.

And then he tried GitHub Copilot on a Saturday afternoon, almost as a joke. By Sunday evening he had rebuilt a small internal reporting tool from scratch, alone, in a language he hadn’t touched in years. His feedback: “I felt like a developer again. And I understood my team’s problems in a completely different way by Monday morning.”

What AI coding agents actually do

There is an important distinction that gets lost in most AI coverage: there is a difference between an AI coding assistant and an AI coding agent.

An assistant waits for your prompt and suggests the next line. An agent executes. You describe a task in plain language, and it reads your codebase, figures out what to change, writes the code, runs the tests, and reports back. You review the result. You don’t need to know the exact syntax. You don’t need to remember how to configure a build pipeline. You describe what you want and you steer.

Tools like GitHub Copilot’s agent mode, Cursor, and Claude Code work exactly this way. The barrier to participation has dropped so dramatically that the question is no longer “can I do this?” It is “do I want to engage?”

Why leaders should care

There are two reasons this matters for anyone in a CTO, CIO, or senior technology leadership role.

The first is credibility. When you ask your engineering teams why something is slow, complex, or expensive, you are usually dependent on their translation. They simplify. They omit. Not because they are dishonest, but because explaining technical trade-offs to someone who hasn’t touched a codebase in years requires a lot of abstraction. If you have recently struggled through a small feature request with an AI agent by your side, you have a much sharper antenna for when the explanation doesn’t quite add up.

The second reason is speed of thought. Strategy today is not just a sequence of slides; it is increasingly a sequence of working prototypes, internal tools, and integrated systems. Leaders who can assemble a small working demo, or at least meaningfully interrogate one, think faster and decide better. Not because they become developers again, but because they remain technically fluent.

The “I’m not a developer” defence

The most common objection I hear is some version of: “That’s not my job anymore.” And that is fair. Nobody is suggesting the CIO should be reviewing pull requests. But there is a difference between doing the work of a developer and staying close enough to the craft to lead it well.

Carol Dweck’s research on growth mindset is useful here. The fixed-mindset version of leadership says: “I have moved beyond this.” The growth-mindset version asks: “What can I still learn from engaging with this?” AI coding agents lower the cost of that engagement to almost nothing. An hour on a Saturday afternoon is enough to close a credibility gap that has been quietly widening for years.

And honestly, there is something else at play. The engineers on your team notice whether their leader can engage with their reality. Not superficially, but genuinely. That noticing changes the dynamic in the room when you sit down together.

Where to start

You don’t need to install anything complicated. Open GitHub Copilot in VS Code, describe a small problem you actually care about solving, and follow the agent’s output. Correct it when it goes wrong. Ask it to explain what it did. Push back when it misunderstands the requirement.

You’ll make something imperfect. That’s fine. The point is not the output. The point is the re-engagement with the texture of the work, with the kind of thinking that your teams do every day.

The competitive edge doesn’t come from the code you write. It comes from the understanding you rebuild.

So: when did you last write something that actually ran?