4 minute read · General information
Describe the job to be done
Begin with a task people recognize: what arrives, what needs to happen, where the work slows down, and what a good result looks like. Avoid starting with a model name or a promise of automation.
A clear description makes it easier to compare options. A consistent checklist, a process change or a better search experience may be more appropriate than a generative system.
Understand information and consequences
Ask what information the task uses, who is allowed to see it, how often it changes and what happens if an answer is wrong. The level of review should reflect the possible impact, not the novelty of the technology.
If a task depends on inaccessible, incomplete or disputed information, introducing AI will not remove that underlying problem.
Test a small, reviewable step
Choose a limited trial with a person checking suggestions and a simple way to record when the result helps or fails. Decide in advance who owns the trial and how it will stop.
The useful outcome is not an AI demonstration. It is evidence about whether a defined task can be improved responsibly.
This article provides general information and perspectives, not professional advice. Evaluate technology choices in your own organizational and regulatory context.