Using AI Without Losing The Plot
Using AI as a tutor and critic, with a concrete way to understand and check the code it changes.
Good AI makes your thinking more visible, not more outsourced.
Ishu D. Garg; edited with AI assistance.
The Bad Workflow
Paste a question, accept the answer, make it look nice. It feels productive, but the mind may barely have touched the problem. The risk is not the tool itself; it is hiding the struggle you meant to practise.
The Loop I Trust
Try this loop: guess, ask, compare, verify, re-solve. Write your rough answer first. Compare it with the model's, check the facts or code, then close the AI and solve a nearby problem.
Good Uses
Ask for a hint before a solution, a counterexample to your claim, or the weak point in your answer. Practice questions, interview simulations, code review, and explanations at different levels can help too. Keep a mistake log and check the sources.
Understand the Change, Not Just the Answer
With code, I want to see what a change does, how it fits into the rest of the system, and what could go wrong. A diagram or a before-and-after example can help more than another page of explanation.
Try it on a small change: ask for the input, the output, and one expected failure. Then run those cases and explain the result in your own words. Also ask whether an existing tool or library already solves the problem; more generated code is not automatically more progress.
Where It Can Make You Worse
A fluent answer is easy to mistake for your own understanding, and a borrowed explanation can flatten your voice.
The Student Rule
If the task is training, bring AI in after your first attempt. If it is drudgery, bring it in earlier—but review the result yourself.
The Question I Keep
The question I keep is simple: what can I now see that I could not see before? A filled page is not the same thing as a learned idea.