Claude Code is changing the developer's job: agents now write the code, not people

What does it mean to be a developer in 2026? Three years ago the answer was obvious: a developer is someone who writes code. Today that line has blurred. More and more people build working programs without typing a single line by hand, and experienced engineers spend their workday not writing code but managing agents.

That is the subject of a recent episode of The Vergecast. The hosts invited Paul Ford, a writer, entrepreneur and one of the most thoughtful technology commentators of the past two decades. Ford has long tracked how programming is changing: back in 2015 he wrote the famous article "What Is Code?", which filled an entire issue of Bloomberg Businessweek.

The starting point of the conversation was Claude Code, an Anthropic tool that lets you delegate programming tasks to AI agents straight from the terminal. Since it appeared, more and more people have started calling themselves "coders", even people who have never written software professionally. Claude Code takes an instruction in natural language and works out how to carry it out on its own: it creates files, fixes bugs and runs tests.

But it is not only about newcomers. The main shift is in how experienced developers work. Professionals now spend less time writing code directly and more time setting tasks, checking results and coordinating several agents at once. It resembles the move from craftsman to site foreman: you are still responsible for the result, but your hands are busy with something else.

A natural question follows: what is left for the human? If agents handle the routine, such as writing functions, debugging and refactoring, where does the human's value lie? Ford argues it lies in understanding context, in the ability to ask the right questions and in accountability for the consequences. An agent does not know why a function is needed or who will use it. A human does.

Another worry is what this means for new developers. The path into the profession used to run through writing code: a lot of code, with mistakes and debugging. That experience built intuition. If a newcomer starts today by managing agents right away, they will lack the foundation that makes it possible to understand why an agent got something wrong. This is a potential systemic problem for the whole industry, and no one is addressing it yet.

Claude Code and similar tools make programming more accessible, and that is good. More people can build tools for themselves and others without spending years studying. But at the same time this creates an illusion of competence: a person sees a working program and thinks they understand the technology. The gap between "I wrote a prompt" and "I understand what is happening" is huge.

Ford frames this as a question of trust: whom do we trust, code we understand or code that works? When responsibility is blurred between human and agent, finding the cause of a failure becomes fundamentally harder.

The future of development is neither the end of the profession nor its continuation in the old form. It is something new that does not yet have a precise name. And as always with technology, how good that new thing turns out to be depends largely on how deliberately we manage the transition.