AI is an excavator, not a shovel — it does not remove the need for someone who knows what they are digging, it changes how much one person can move in a day and how badly a mistake scales.
This is a short personal piece rather than a technical guide. It is the analogy I use when clients ask how I work with AI on Drupal projects, and why I do not think it replaces the judgement they are actually paying for.
The analogy
Say you are putting a pool in your back garden. Before anything else you call the city, because there are utilities under there and finding them with a bucket is the wrong way to find them. Then you mark the area and start.
Now you have a choice. Ten people with shovels will get it done in about five days. One skilled operator with an excavator will do the heavy work in three or four hours, and then two or three people finish the edges by hand.
Both work. The second is faster by an order of magnitude — and it carries a risk the first one does not. Put someone inexperienced in that cab and you can breach a gas line, undercut a foundation, or hit something that turns a weekend job into an insurance claim. The machine does not make the operator careful. It makes the operator faster, in whichever direction they were already heading.
Why this is the right analogy for AI
Every property of the excavator holds:
- Leverage cuts both ways. A good instinct arrives sooner. So does a bad one, at the same speed and larger.
- The survey still comes first. Calling the city is the equivalent of understanding the codebase. Skip it and speed is the problem, not the solution.
- Finishing is still by hand. The machine does the bulk. Edges, tolerances and the bits that have to be right are human work.
- Skill is in knowing the limits. An experienced operator knows what the machine cannot do in a tight corner. That knowledge is the job.
What this means in practice
The place AI has changed my work most is not writing code. It is the reading — getting oriented in an unfamiliar module, checking what an API actually does, drafting the boring half of something so the interesting half gets attention.
And the discipline that goes with it is simple: verify against the source, not against the answer. A confident wrong answer about a Drupal API looks exactly like a confident right one. The file on disk settles it, and checking takes less time than being wrong does.
That is not scepticism about the tool. I grew up around machines, and the thing you learn is that you build a relationship with one — you learn its rhythm, where it is strong, where it will let you down. Tools become extensions of the people who take the trouble to learn them.
Common questions
Does AI replace a Drupal developer?
It replaces some of the typing. It does not replace deciding what to build, understanding an existing system, or being accountable when it is wrong.
Should I be worried about AI-generated code in my project?
Worry about unreviewed code, whatever produced it. The question to ask a supplier is what their review process is, not what tools they use.
Related
The longer version of this argument, applied specifically to Drupal work, is in how AI changes what a Drupal developer does. If you would rather see the output than the philosophy, the portfolio is the shorter route, and a direct question is shorter still.