This article was written in collaboration with an LLM (ideas belong to the author).

If you are a computational researcher, the recent rise of AI-enabled “vibe research” has probably not done much for your already existing imposter syndrome.

After all, if programming was the main skill you brought to the table, these are difficult times indeed. The ability to write code, debug it, build pipelines, run analyses, and even implement fairly sophisticated models is rapidly becoming less scarce. An LLM can now do much of this work—often faster than you can.

And this creates an uncomfortable question for the computational researcher: if coding was the thing you were being paid for, what exactly do you bring to the table that an LLM does not?

There is an even more uncomfortable version of the question. If you are doing roughly the same amount, and producing roughly the same quality of work, as you were before AI became ubiquitous, then perhaps you really do have something to worry about. The productivity gains are no longer hypothetical. If the tools have become dramatically better but your output has not, the feeling of being an imposter may not be entirely irrational.

But perhaps this is also the wrong way to think about the transition.

The value of a computational scientist was never really in the ability to type Python or MATLAB code. Code was merely the instrument. What mattered was knowing what to compute, why to compute it, how to formulate the problem, what assumptions are being made, whether the result is meaningful, and where the analysis is fooling you.

And this distinction may become much more important in the post-AI world.

If AI makes producing analyses cheap, then focus and depth become the scarce resources. In a world where almost anyone can generate a figure, fit a model, run a statistical test, or produce a plausible manuscript, the ability to identify an important question and pursue it deeply becomes increasingly valuable.

This may also create a paradox.

Vibe research could dramatically increase the volume of scientific work without producing a corresponding increase in scientific understanding. If the marginal cost of generating an analysis approaches zero, we should expect an explosion of experiments, figures, preprints, papers, benchmarks, and ultimately noise. The amount of scientific information may increase enormously while the signal-to-noise ratio decreases.

In other words, AI may make junk science cheaper before it makes great science easier.

That has consequences for how scientific labour is valued. Traditionally, a PI could hire a PhD student or postdoc because they brought a bundle of capabilities that were difficult to obtain elsewhere: technical expertise, domain knowledge, intellectual autonomy, persistence, and the ability to turn vague questions into working experiments. The computational component of this bundle is now increasingly available on demand.

So the uncomfortable question for the purely computationalist becomes even sharper: what does a PhD or postdoc provide that an LLM does not?

The answer cannot simply be “better coding.”

It has to be something closer to scientific judgment.

The ability to recognize an interesting problem. To know when an apparently beautiful result is actually trivial. To understand the literature deeply enough to see what everyone else has missed. To design an experiment whose outcome would genuinely discriminate between competing explanations. To sit with a problem for months without needing an immediate reward. To develop taste.

These are much harder things to automate.

Perhaps science, like many other domains, will eventually self-organize around this new reality. As AI makes execution abundant, problem selection, attention, judgment, and depth may become the new bottlenecks.

This suggests a different way of thinking about imposter syndrome in the age of AI.

If your identity as a computational researcher rests primarily on being able to produce code, then yes—the ground beneath you is moving.

But if your value lies in knowing what deserves to be computed in the first place, then perhaps the AI revolution is not the end of your career. It is a revaluation of what your career was actually about.

The post-AI world may not reward those who can produce the most.

It may reward those who can decide what is worth producing.