AI is changing the way we work as developers.
We are moving from writing code to writing specifications. Instead of implementing features ourselves, we rely more on agent modes and tools like Claude Code, Codex, or Gemini CLI.
This works incredibly well. Productivity increases, repetitive work disappears, and we can ship features faster.
But there is something I’ve started noticing.
Developers are not necessarily learning more.
We are just doing the same things faster.
AI Cannot Replace Understanding
When you rely heavily on AI tools, something subtle happens.
You stop exploring new solutions yourself.
You stop experimenting.
You stop learning new paradigms.
Many developers simply apply the same solutions they always used, just faster because AI writes the code.
When a new approach or development practice appears, it often does not get adopted well because the developer does not really understand it.
AI can generate code.
But it cannot give you knowledge you never built.
The Knowledge Gap
A simple example.
I am not an expert in:
- load testing
- monitoring
- observability
I know these areas are important. But if I do not understand the concepts, how am I supposed to apply them correctly?
Even with the best prompts and agents, the result will still be shallow.
AI helps execute ideas. It does not replace learning them.
Where This Shows the Most
I notice this especially in two areas.
- testing
- accessibility
Testing
Recently I have heard things like:
Why write tests if AI generates almost perfect code?
or:
We run Playwright tests with AI after the feature. That is enough.
But testing is not about validating one run. It exists to prevent regressions, catch old bugs, and protect systems as they evolve.
AI can generate tests.
But testing strategy still requires understanding.
Accessibility
Accessibility has always been ignored in many projects.
AI has not changed that.
You can add an accessibility agent, but accessibility is not just adding attributes. It requires understanding how assistive technologies behave, how screen readers navigate, and how keyboard interaction works.
Without that knowledge, accessibility stays superficial.
The New Developer Loop
Something interesting is happening.
Developers spend more time writing Markdown specs for AI than learning new technical concepts.
Instead of experimenting, breaking things, and learning through failure, we optimise prompts and agent workflows.
But not necessarily ourselves.
I Am Also In That Loop
To be fair, I am also part of this.
AI tools are powerful and it is easy to rely on them.
But recently I started asking myself a simple question.
Am I actually improving?
Or am I just getting faster?
So I am trying to go back to learning new concepts, experimenting more, and understanding systems deeper.
And only after that, delegating to AI.
Ask Yourself This
If AI makes you 10x more productive, ask yourself:
Are you doing something you would never do if you had infinite time?
Are you exploring areas you previously did not understand?
Or are you just shipping the same things faster?
And one more question.
If your productivity increases 10x, do you now have 10x more time?
More time to learn, exercise, spend time with family, or build deeper knowledge?
If the answer is no, then what is AI really doing for you?
Is it helping you become a better developer?
Or just a faster version of the same developer?
Top comments (0)