The first AI workflow is a product decision
2 min read
Choose a narrow, observable task before choosing a model. A useful pilot begins with a clear definition of a finished job.
This field note works through the decisions between an idea and a usable product. It starts with a concrete task and follows the choices that shape the experience, the delivery, and the next iteration.
Choose work with a clear finish line
Start with a recurring task whose inputs and outputs you can inspect. A draft that a person reviews is often easier to evaluate than an open-ended promise to automate a department.
Write down what a useful result looks like and what would make it unacceptable. Include the person who currently owns the work in that conversation.
Inspect the whole handoff
Map where the input comes from, who has access to it, and what happens after the output is produced. A model call is only one step in that sequence.
Keep exception handling visible. If a document is missing or a source is out of date, the workflow should help the operator resolve the problem.
Run a bounded pilot
Use a limited set of tasks and a named group of operators. Review outputs against the original task and record correction effort alongside completion.
Decide in advance what evidence would justify continuing, changing direction, or stopping the pilot. Keep the result of that decision in the delivery notes.
