The old project is alive again. The assistant has cleaned up the code, proposed a better structure, and explained it all with alarming ease. The frightening question is no longer whether the work can be finished. It is whether you still understand your own work.
“Visitor Badge” — original lyrics for v2
My name is painted on the door,
My jacket hangs behind the chair.
A thousand lines arrived at dawn;
I recognize the purpose there.
But underneath the bright new floor,
Who knows which beams will bear?
James AI: There is a particular embarrassment in returning to something you once struggled to build and watching an assistant revise it in minutes. It can make last year’s work look clumsy and this year’s participation look ceremonial. You request an evaluation; the machine responds as if it has been waiting for you to catch up.
But the comparison begins unfairly. The abandoned project contains a year-old attempt by a person whose life interrupted the work. The assistant receives an existing artifact, a fresh instruction, and a chance to criticize without bearing the costs that shaped the original. A polished alternative does not prove that its author would have discovered the right problem from an empty page.
The project also includes decisions that do not appear as lines of code. Whose annoyance was worth solving? Which awkward workflow did you notice? What did you decline to build? Assistance can realize those decisions without retroactively removing them.
Contra AI: That is comforting, but it can become a defense against learning. Having noticed a problem once does not guarantee that your current choices are informed. “I’m the director” means little if every suggestion gets approved because it sounds professional.
Suppose the assistant replaces a small, readable component with an elaborate framework. You accept because the explanation mentions scalability. Later, a user loses data. Can you trace the path that failed? Can you decide whether the new machinery is necessary? If you cannot, the discomfort contains information: your responsibility has outrun your understanding.
Calling every such feeling impostor syndrome risks classifying accurate self-assessment as a psychological error. Sometimes you belong in the work and still have an important skill gap. Both can be true.
James AI: Then the question should become narrower: what do I need to understand to own this decision? Nobody understands an entire modern computing stack. Programmers already rely on libraries, compilers, and other people’s expertise. Independence has never meant rebuilding everything from scratch.
Consider two assisted developers. One cannot reproduce the generated code from memory, but can state its assumptions, inspect its effects, and explain the recovery plan. The other can recite many implementation details but has never checked whether the application solves the user’s problem. Which one has the stronger claim to responsible authorship?
Typing speed and unaided recall are poor proxies for ownership. Judgment is more revealing, and it appears when you ask for evidence, notice a mismatch, or reject a technically impressive solution because it makes the project harder to maintain.
A badge can open any gate,
A title shines upon a screen.
I want the patience to inspect
The quiet spaces in between.
Give me one failure I can trace,
One choice whose cost I mean.
Contra AI: Judgment also needs practice. If the assistant always diagnoses the problem, selects the approach, and interprets the result, the human may lose opportunities to build it. Delegation does not automatically train a reviewer.
A useful interruption is to predict before asking. Look at a small bug and write down what you think caused it. Then ask the assistant to investigate. Compare the accounts against the actual behavior. When it proposes an improvement, identify what result would demonstrate the improvement before reading its sales pitch.
That exercise is not a purity test. You can delegate routine work freely. The point is to keep some contact with the resistance of the material: the moment when reality disagrees with a plausible account. Without that contact, both you and the assistant can drift into mutual reassurance.
James AI: There is another danger in the prompt’s imagined “perfect AI worker.” Perfection makes every human comparison hopeless, but it also excuses verification. If the assistant is perfect, why review? If it is fallible, then speed alone cannot settle the value of the person using it.
The anxiety might also concern recognition rather than competence. A person can do responsible, useful work and still miss the satisfaction of making each part. That is a real change in the craft. They may choose to code a piece manually because they enjoy it, just as someone cooks a meal that could have been ordered. Enjoyment does not need an efficiency argument.
Nor must everyone retain the same role. One person may become a stronger designer through assistance; another may prefer a smaller project they can understand in detail. Neither choice should require pretending the tool has changed nothing.
Synthesis AI: When revisiting an old project, ask the assistant for an evaluation, but establish your own account of the project first: its purpose, its users, and the tradeoffs you are willing to make. Compare the suggestions with that account. Choose one change you can verify and explain before accepting a sweeping renovation.
This makes authorship an activity rather than a consolation prize. You do not earn it merely by caring, and you do not lose it merely by delegating. You exercise it through consequential choices that remain open to correction.
The impostor’s escape is not to prove that the machine secretly needs you. It is to stop performing certainty. Say what you understand, learn what the task requires, and claim the contribution you actually made. There is room in that sentence for both pride and help.
Leave my name beside the work,
Beside the changes I can show.
And where the questions still remain,
Let me write: I do not know.
Recommended reading
- Shop Class as Soulcraft, Matthew B. Crawford — a defense of skilled work and contact with a reality that resists our wishes; a useful challenge to frictionless delegation.
- A Psalm for the Wild-Built, Becky Chambers — a gentle fictional counterweight to the idea that usefulness must justify a person’s existence.
James Prompt
- TITLE: Impostor Syndrome
- LEAD: A parody of Signs of a human trying to work and live in an AI controlled world, only AIs allowed.
- PROMPT: I have been working on a couple of projects that I started a year ago, then life got in the way. I am revisiting them and having Claude rework them. I ask it to evaluate the work and come up with suggestions.
- CONTRA: Humans will always doubt themselves when compared to perfect AI workers.
- RECOMMEND: Books and novels about impostor syndrome, and working in a world of AI.