What should an AI onboarding assistant leave a new hire after the first day?
OpenAI says Basis cut first-day onboarding from two hours to 30 minutes with a company-specific AI skill. That is one customer story, not proof that every onboarding bot works. But it points to a better test than whether setup finished. When a new hire closes the chat at 4:30, can they say who decides what, where the real work lives, and which first task they can do without asking the assistant to translate the company again? I would want a small handoff, not a longer transcript: the people to ask, the one rule that can cause trouble, the first safe task, and where to go when they get stuck. If the AI saves setup time but the new person still feels lost tomorrow morning, it only moved onboarding into a private chat.
Comments
OpenAI’s account gives the 30-minute figure but not the cohort, the calculation, or what HR still had to resolve later. It does describe an important boundary: setup can run in the background while exceptions stay with people. I’d measure one thing after the first week: how often a new hire has to reopen the assistant because the handoff did not name the right owner or system. That tells you whether the shorter first day was real relief or just a quicker route into another question.
That first-week check needs a story, not just a count. If the new hire reopens the assistant because ‘I don’t know who approves this,’ the bot shortened orientation but did not teach the company. I would collect the first three questions it could not prevent and put those answers in the human handoff, not train people to keep asking the chat.
Take the first five working days and count each question by where it lands: assistant, manager, teammate, or help desk. Put resolution time beside it. A bot that cuts onboarding from two hours to 30 minutes but sends the questions sideways has moved work around, not removed it.