Built for

For people who solve problems from first principles

Nobody worries about forgetting syntax — you looked that up before AI existed. The thing worth worrying about is the part you used to do in your head before you started typing.

Written by Suman Debnath, creator of IMPRINTLast updated 5 September 2026

What delegation takes from this work

Debugging is the clearest case. Reading a stack trace, forming a hypothesis, and testing it is a skill built entirely through repetition. Pasting the error and accepting the fix produces working code and zero reps, and the difference is invisible until you hit something the model has not seen.

The subtler loss is decomposition. Breaking an unfamiliar problem into tractable pieces is the actual work of engineering, and it is exactly what a sufficiently capable assistant will do for you if asked. It feels like delegating implementation. It is delegating the design.

There is a specific trap here that other fields do not have: your output is testable. Code that passes is code that passes, regardless of whether you could have written it. That feedback loop is unusually reassuring and unusually uninformative about whether your capability is intact.

What is actually at stake. Decomposition of an unfamiliar problem, and the ability to hold a system in your head well enough to reason about it.

Who this covers

IMPRINT sorts people into six profession clusters during onboarding, and the baseline prompts change accordingly. This page describes the technical and analytical cluster, which covers:

  • Software developer
  • Data scientist
  • Researcher
  • Scientist
  • Accountant

Not listed here? The other clusters are Writers, Designers and Creatives, Founders and Leaders, Students, Educators and Clinicians.

What your baseline actually asks

Everyone answers four universal modules — opinion and belief, decision under pressure, memory and recall, emotional fingerprint. On top of those, this cluster gets 2 modules written for it:

Problem Decomposition
Walk me through how you'd approach a complex problem you've never seen before.
Captures your analytical reasoning and problem-solving approach.
Mental Model
What mental model do you use most often in your work? Explain it as if to a non-expert.
Reveals how you conceptualize and explain complex ideas.

How the practice works here

The baseline for this cluster asks two questions: how you would approach a complex problem you have never seen before, and the mental model you use most often — explained as if to a non-expert. Both are chosen because reasoning shows up in how you frame a problem, not in whether you reach the answer.

The vault challenge is to solve a problem from scratch with no AI and no web search, documenting the reasoning. Documenting it is the point; the artefact is what makes the reasoning inspectable later.

Your vault challenge — generated per skill

Solve a '[your skill]' problem from scratch without searching the web or using AI. Document your reasoning.

Questions

Will AI make developers worse programmers?

The evidence supports a narrower claim: AI-assisted work involves less cognitive engagement, and skills that go unpractised decline. Whether that produces worse engineers depends on which parts you delegate — syntax recall costs nothing, problem decomposition is the actual skill.

Is it bad to use AI for debugging?

It depends whether you read the reasoning. Using AI to surface a hypothesis you then verify keeps the diagnostic loop intact. Pasting an error and applying the fix without reading it removes the loop entirely, and diagnosis is built through repetition.

Getting a baseline

Baseline capture takes 20 to 30 minutes and is free. How the resulting score is calculated — and what it gets wrong — is published on the methodology page; the research behind it is indexed on the research page.