2 Employee Onboarding AI Agents, Live in Production.

Onboarding documentation has a predictable decay. A handbook is written once, policies change underneath it, and the version a new starter receives contains statements that are no longer true — stated with the same confidence as the ones that are. Because nothing records where each statement came from, nobody can audit it, so the drift is invisible until a new joiner acts on something outdated. Task documentation fails differently: the person who knows how a process actually works has done it a thousand times, so the write-up omits the steps that have become automatic. What reaches the new starter is a procedure with three invisible gaps, and they fill them by interrupting a colleague, which is the cost the document was meant to remove.

These agents make provenance explicit. Onboarding Handbook Generation builds a role-specific handbook from your actual policies, with every statement traced to a source and the gaps marked rather than filled in. Marking a gap is more useful than covering it: an acknowledged gap gets closed, and a plausible invention gets believed. Training Documentation turns how a task is actually done into something a new starter can follow, with the steps that only exist in someone’s head written down and the ones nobody could verify marked as such. That marking matters for the same reason — an unverifiable step in a procedure is a place where the documentation and reality may already have parted, and flagging it is the honest output.