The classic audit trail answers an accountant's question: who touched this number and why. The automation-era version answers a broader one: when software acts on your behalf, what exactly did it do, and on whose authority? A complete trail for an AI-prepared action records the source data the AI read, the draft it produced, its stated reasoning and risk level, the human decision (approved, edited, declined), the person who made it, and the result the target system returned.
Two properties separate a real trail from a log file. Completeness: declined and edited actions matter as much as approved ones, because they document judgment being exercised. And legibility: a trail a new bookkeeper can read is an asset; a JSON dump only the vanished consultant understood is not.
Small business example: a customer disputes a payment reminder from March. The owner opens the trail: the reminder cited invoice #1082, was drafted from the QuickBooks aging report on March 4, edited to soften the tone, and approved by the office manager at 9:12. The dispute resolves in one email, with dates. That is the quiet value of the trail: not catching villains, but ending arguments.
See the term in practice
Connect your tools read only and watch the approval queue prepare real actions from your own context. Nothing sends until you approve it.