Your AI Intern Just Started. Who's Supervising It?

The easiest way to think about AI at work is as a new intern. It's fast, never tired, eager to help, and sometimes confidently wrong. Nobody would hand an intern on day one the keys to every file and let their work go to customers unchecked. AI deserves the same supervision.

I use AI in my own work every day, and it's at its best when I treat it like a sharp new hire: fast, useful, and always reviewed.

Give it the right work

AI is good at first drafts, summaries, rewording, organizing notes, and answering "how do I" questions about common software. It is weaker at anything that must be exactly right the first time: prices, legal terms, technical settings, and facts about your specific customers.

• Good jobs: drafting an email reply, summarizing a long thread, turning meeting notes into action items, cleaning up a spreadsheet formula.
• Jobs to supervise closely: anything with numbers, dates, contracts, or advice a customer will rely on.
• Jobs to keep away from AI: decisions about people, anything regulated without review, and anything you couldn't explain if a customer asked how it was decided.

Check its work

AI can state something false with complete confidence. Treat everything it produces as a draft from a smart new hire: read it, check the facts that matter, and own what goes out. Make it a team rule that a person is responsible for anything AI helped write.

Watch what it can see

Tools like Microsoft 365 Copilot can only see what the person using them can see. That sounds safe until you remember how much most people can see: old shared folders, team sites nobody owns, the spreadsheet someone shared with everyone years ago. Before switching on an AI assistant that searches your files, tidy up who has access to what. Otherwise it will happily surface the payroll file in answer to an innocent question.

Pick the tools, don't let them pick you

AI is now built into email, browsers, accounting software and phone apps, often switched on by default. Rather than a free-for-all:

• Choose one approved AI tool on a business plan, so your data isn't used to train public models.
• Turn off AI features you don't need in the software you already use.
• Write down, on one page, what can and can't be put into AI tools.
• Revisit the list every few months. This is moving fast.

The payoff

Used this way, AI really does save time: the hour spent writing a proposal becomes twenty minutes of editing a good draft. The businesses getting value from it aren't the ones using the most AI. They're the ones supervising it well.