You can't debug what you don't understand.

AI has given everyone the ability to build custom software in plain English. No computer science degree required. No bootcamp. Just describe the problem and iterate until it works.

For mid-market operators this is a genuine opportunity. Custom internal tools that used to require a developer and weeks of work now take hours. That productivity gain is real and it matters.

But there is a cost that is not showing up in the productivity metrics yet.

When coding only requires English and a problem statement, you never build a mental model of how software actually works. You learn to operate a system without understanding it. That is a completely different skill. And when something breaks, asking the same tool that wrote the broken code to fix it is not debugging. It is a retry loop with no exit condition.

There are two reasons for optimism though. First, people who build their own tools in plain English may be forced to learn how the machinery works when things break. Necessity is a powerful teacher. This could drive broader software literacy rather than reduce it. Second, AI will likely get better at debugging its own code over time. If so, this problem could be shorter-lived than it appears. Hard to know.

In the meantime, two questions for every engineering and technology leader. Can your developers explain the code they ship? Can they find a bug with or without AI? Code that is not understood cannot be maintained. Code that is not maintainable is more of a business liability than an asset.

Note: The New Stack discloses that its parent company Insight Partners is an investor in Octopus Deploy, the source of some of the hiring data cited in this article.