Write an AI Tool Note That Another Developer Can Reproduce

“It worked for me” is a weak handoff for an AI tool. A teammate needs the input, the settings, and a clear account of what actually ran. A short experiment note can provide that without turning into a full review.
Separate discovery from evidence
A directory such as ewiki can help you find a tool or understand its stated purpose. Treat that as a starting point. Availability, pricing, and output behavior still need checking in the service you evaluate.
Record the run boundary
Input: owned sample, saved unchanged Settings: record the visible values Action: setup only OR generation submitted Output: file saved OR no output produced 
This tiny record prevents an upload walkthrough from becoming an accidental quality claim. Include a failure too: a rejected file, a confusing permission, or an output you could not export.
Make screenshots useful to readers
Explain the important setting in text instead of hiding it inside an image. The W3C alt decision tree helps decide what information an image alternative should carry. A screenshot caption can identify the step while the surrounding prose preserves the actual instructions.
Finish with the next reproducible action, such as testing a second file with the same settings. That gives a teammate something concrete to verify and keeps one promising example from standing in for a broad conclusion.
Add the date and the interface language to the note. A renamed control is easier to locate when the reader knows which version of the workflow your screenshot represents.