Make your first AI experiment easy to undo
Practice on a copy, compare one change, and keep a route back to the work you trust.

Watch the 42-second reversible-experiment lesson
You do not need to risk your best work to start
If you have never used AI for a task you care about, starting can feel like handing over the whole process. You might worry that it will rewrite your voice, lose an important detail or leave you with a version you cannot repair. There is a smaller first step: make the experiment easy to undo.
Imagine a fictional maker who has written instructions for folding a paper house. The goal is not to replace the maker's experience. It is to see whether one paragraph can become clearer. That is a building decision: limit the change so you can understand what helped and what did not.
Create a practice copy with a clear boundary
Begin with a short, harmless sample you wrote yourself. Keep the original unchanged and make a separate practice copy. Give the two files names you can distinguish, such as “paper-house-original” and “paper-house-practice.” Open both once to confirm that you can still find the original.
Do not use customer records, private employer documents or confidential methods for this first exercise. A copy protects the original file from editing; it does not make sensitive information safe to upload. Choose material that is appropriate to share with the tool in the first place.
Now ask for one change: “Suggest a clearer version of this paragraph. Keep the order of the folding steps. Do not add materials or measurements. Show the proposed paragraph separately.” Treat those words as instructions to review against, not as a technical guarantee that the tool cannot make other changes.
Compare before you replace
Read the proposed paragraph beside the original. Did the tool remove a necessary fold? Did it introduce a measurement you never supplied? Did it replace a familiar word with language a beginner might misunderstand? You are checking a specific piece of work, not deciding whether AI is generally good or bad.
For example, the original might say to fold along a marked line before attaching the roof. A shorter rewrite that skips the marked line is not an improvement simply because it uses fewer words. Keep the useful wording, reject the missing step and write down your reason. If nothing improves, keep your original. An experiment can be informative without producing a replacement.
Before you finish, practice returning to the original in this harmless example. Do not assume every tool has the same version history or undo behavior. If you cannot identify a dependable way back, do not use that workflow to overwrite important work.
Save the decision as part of your portfolio
Jakob Nielsen's usability heuristics include user control and freedom, with clear exits and support for undo and redo. The NN/g article was published April 24, 1994 and last reviewed January 30, 2024. These are broad design rules of thumb, not evidence that this particular AI exercise has been experimentally validated.
The exercise here applies that principle to a beginner's working habit. Save the original, the proposed change and one sentence explaining what you accepted or rejected. In your knowledge portfolio, that sentence helps show your contribution: you recognized which detail mattered and made the final choice. Label the AI assistance rather than presenting its rewrite as entirely your own work.
This is not a promise of legal protection, licensing income or future employment. It is a practical way to build experience while retaining evidence of your decisions. Start with one paragraph today. Keep the original, try one change and demonstrate that you can return to the version you trust. Confidence can begin with knowing how to stop.