Record the explanation as well as the action
Ask the operator to show the job from preparation through the final check. Capture the work area, tools, parts, and the correct result. The recording should make part positioning and important hand movements visible.
Ask what a new operator is likely to miss. Experienced people often make small decisions without describing them: how they recognize a wrong part, why a sequence matters, or when they stop the job. Those explanations make the resulting instruction more useful than a list of actions alone.
Give AI the current source material
Upload the demonstration with the existing instruction, relevant drawings, and available CAD. In Seneca, those inputs support a draft procedure with steps and visual references. Keep a record of the source so reviewers can revisit the demonstration.
Conflicting files need review. If the recording shows a newer fixture than the written document, the draft should not silently combine both methods. Identify the approved version of the job and resolve the difference with the process owner.
Review each step for action, context, and check
For every step, ask three questions: what should the operator do, what detail makes it correct, and how will they know it is complete? A short annotated image or video segment can explain a position more clearly than a long paragraph.
Keep checks concrete. “Check the assembly” gives little guidance. A reference showing the correct orientation and the area to inspect gives the operator something observable. Any numerical requirement must come from the approved engineering or quality source, not from an AI guess.
Approve, publish, and update the procedure
Route the draft through the teams that own the method and its checks. Production, quality, and safety may review different parts of the same instruction. Record the approved version before using it for training or floor guidance.
Make the current instruction easy to open at the job, for example with a QR code. Give operators a way to send feedback beside the step that needs attention. When the process changes, capture the revised method, review the affected steps, and update the connected training rather than recreating unrelated copies.