There is a way of using these tools where you get faster and stay yourself. There is another where you get faster and slowly stop being anyone in particular, and from inside, the two feel identical.
I catch myself on the second path more often than I would like to admit. The first plausible answer is almost always fine, and “fine” is the whole problem. I use these tools every working day, on backend and infrastructure work, so this is a practice I need rather than a warning I am issuing.
Output and judgment are not the same thing
Almost everything written about AI treats them as one thing, and that is where the damage starts.
Output is the artefact: the code, the draft, the plan document. Producing it is labour, usually the slowest part, and handing it over is a straightforwardly good trade. Nobody’s professional identity should rest on having typed something.
Judgment is deciding what should exist, whether the thing in front of you is any good, and what to do next given everything you know that is not written down anywhere. It is slow because it needs you to hold a lot of context and be willing to be wrong in front of people.
Hand over the first and you get leverage. Hand over the second and you have given away the part that was yours.
Why the second one happens by accident
Nobody decides to stop making decisions. It erodes, because the tools are good enough that accepting the first plausible answer is almost always fine.
Ask for an approach and you get a reasonable one. Ask what to build next and you get something defensible. None of these are wrong. They are the median of everything the model has seen, which makes them the median of what everyone else in your position would have done.
Median answers do not feel like a compromise in the moment. They feel like competence. Nothing goes wrong. Nobody objects in review. You only notice the cost in aggregate, over months, when everything you have produced is reasonable and none of it is distinctive.
The muscle that atrophies is the one that says this is fine, and it is not what I want. That sentence is expensive. It means more work, and you have to justify it to someone. The tools make it very easy to never say it again.
Two signs I watch for in myself, because neither feels like a problem from inside: I cannot explain why a piece of my own work is shaped the way it is without re-reading it, and I cannot date my last real disagreement with a proposed approach.
Where the value actually sits
Consider what does not come out of a prompt.
Knowing which problem is worth solving. Models are excellent at solving the problem you hand them. They have no view on whether it is the right problem, because that depends on things that are not in the conversation: what the business is trying to do, which customer complained loudest and whether that customer matters, what the team can realistically maintain.
Taste that survives an argument. Anyone can have a preference. Being able to say why one structure will hold up and another will not, in terms of the specific constraints of this system, comes from having been wrong before, in this codebase, with these people. There is no shortcut to it.
Willingness to go against the median. Every genuinely good decision I have made professionally looked slightly unreasonable at the time to at least one person in the room. Median answers never look unreasonable, which is precisely their limitation and why they are so comfortable to accept. Knowing when that disagreement is worth spending and when it is not is its own skill.
The drive to actually move. No tool has ever wanted anything. It will not decide today is the day, and it will not push through the boring middle of something that stopped being interesting three weeks ago. That stays entirely on you, and it is most of the reason anything ever gets finished.
The practice that keeps it intact
The grand version is a career that stays distinctive. The ordinary version is three habits, in order of how much they matter to me:
- Position first – Write down what you think the answer is before you ask, because thirty seconds of a view turns the reply into a proposal you evaluate instead of an answer you receive.
- Draft always – Treat the first answer as a draft every time, because the only defence against a gradual process is a policy rather than a judgment call made when you are tired.
- Hard part – Debug something by hand now and then, because the understanding you get from struggling with a system is what your judgment is made of, and you cannot buy it back later.
The third is the one I have had to be most deliberate about. There is no single day on which skipping it is the wrong call, and that is exactly the problem.
To be fair to the other side: plenty of decisions do not deserve the attention, and not being able to explain a choice is sometimes fine. If it is happening on the ones that matter, it is a signal, and it is the earliest one you will get.
Friction was doing quiet work
These tools removed almost every source of friction between having a vague idea and having a finished artefact. Friction, annoying as it was, forced you to decide whether the idea was worth the effort before you spent it. Remove it entirely and you produce far more, with far less deliberation about whether any of it should exist.
Being intentional is the replacement. It has to be deliberate now, because it is no longer enforced by the cost of doing the work, and nothing prompts you to make the adjustment.
There is an asymmetry underneath this. Concentrating your entire value in something highly automatable, with no distinctive judgment behind it, is a position with poor odds, and poor odds sized too large are what actually ruin people, never any single bad outcome.
A year from now, the part of your work that is yours will be whatever you refused to accept on the first pass. Whatever you build, that is the part worth protecting: what you decided, what you understood the hard way, and what you were willing to push for when the median answer said otherwise.
The tool answers faster than you can think; only you can decide whether the question was worth asking. When did you last say “this is fine, and it is not what I want” out loud?