Hire for AI skill in the office

Every business runs on people who do the physical work. Someone digs the ditches, someone cooks the food and someone takes out the trash. That's the actual work, and the people who do it well have real skill.

Plumbers and electricians are gods, and so are carpenters, roofers and bricklayers.

The office is different. Anyone you hire for a desk should be a B or better at AI. Hire the B and work on turning them into an A, and once you've done that, you have a real employee.

Telling a B apart from someone who typed "AI" onto a resume isn't hard. Watch them work, ask how they checked what the AI produced and look at what they've built.

The office bar

For an office job, I want someone who uses AI to automate, to organize and to create. Asking a chatbot a trivia question once doesn't count.

The job might be answering phones, and that's fine. Even so, a person at a desk isn't washing the car or taking out the trash between calls. That dead time should go to building automation and developing ideas for the sales funnel. Dead time at a desk is build time.

Are employers really asking for AI skills?

Yes, and they're paying for it. In 2024, Microsoft and LinkedIn surveyed 31,000 office workers in 31 countries. Two out of three business leaders said they wouldn't hire someone without AI skills. In the same survey, 71% said they'd take a less experienced candidate with AI skills over a more experienced one without them.

Lightcast, a company that tracks job postings, analyzed more than 1.3 billion of them. Postings that asked for AI skills offered 28% higher pay, close to $18,000 a year more.

Few people are skilled at it yet. In September 2025, Pew Research Center found that 21% of U.S. workers said at least some of their work was done with AI. Microsoft's 2026 survey covered only people already using AI at work, and it placed just 16% of them in its top group.

That means a truly skilled user is still hard to find.

What does a B or an A look like in an interview?

The one thing to listen for is whether the candidate asks AI questions or builds with it. The table below is one way to sort what you hear. It's a sample to think with, not a scorecard.

Microsoft found the same split: its top users were more likely to decide up front whether AI or a person should handle a task, 53% against 33%.

How do you test a candidate's AI skills?

The first way is to use Claude to build a short test made of real AI tasks from the job. The second is to ask candidates to bring an AI portfolio to the interview, or to send one shortly after. It can hold scripts, skills, small programs and work they've made with generative tools.

Ten minutes of watching someone work will tell you more than their resume.

Keep it the same for everyone: the same task, the same questions and the same time limit, all tied to real duties of the job. Tell candidates that AI helped you build the test. Claude builds it, and you score the work and make the call. Claude never scores or ranks candidates. If a candidate asks for an accommodation, such as extra time or another format, work with them. The same rules apply to a written job-fit questionnaire.

What AI interview questions should you ask?

These are suggestions, so tie each one to the real job, keep it about the work and ask every candidate the same set.

1. "Walk me through the last thing you automated with AI. What did it replace?"
2. "Show me a prompt or saved instruction you use every week. Why is it written that way?"
3. "Tell me about a time the AI got something wrong. How did you catch it?"
4. "Here's a long vendor email thread. What would you give to AI, and what would you check?"
5. "What part of this job would you never hand to AI?"
6. "What's an AI tool you stopped paying for, and why?"

What goes in an AI portfolio?

A folder or a link is enough. Here's what would impress me: batch files and automations, shortcodes, widgets and small apps built from real design files, and scripts. Scripts in Excel or Google Sheets, or anywhere else in Google Workspace or Microsoft 365, would be huge. Whatever they bring, ask how they checked the output before anyone else saw it.

Tell candidates not to include a former employer's confidential work, or anything that belongs to that employer's customers.

Why does AI skill matter for an office hire?

This point matters more than anything else in the post.

The most skilled AI users can sit down for 10 hours and finish a month's worth of work. I can now do a month of traditional work in one 10-hour day, though only on the admin and sales-funnel side. I still have to take out the trash.

Picture three of those people in your office, then five, then seven. Compare them with someone who works the old way, reading every email and typing every reply.

"Dead time at a desk is build time."

What could a 10-hour AI day look like?

Here's a made-up example, not anyone's real schedule: a strong AI user working as the office manager at a small business with two locations.

In 2025, Anthropic studied real Claude conversations and estimated that Claude cut task time by about 80%. For invoices, memos and other documents, the estimate was 87%. Anthropic says plainly that these are Claude's own estimates, not timed results. Earlier controlled studies found much smaller savings.

Notice what's missing from that schedule: someone still has to take out the trash.

Talk, don't type

Typing is the biggest waste of time in an office, so if someone on your team can't type 70 words a minute, they should talk to the computer instead. The average typist manages about 52. I want to see my people talking to their computers, not typing. The Do post Stop typing: talk your first drafts shows how to set it up in about 30 minutes.

Give them the right seat

Put your office team on a Claude Team or Enterprise account, and give each person the seat that fits their work. As of October 20, 2026, a Team Standard seat costs $25 a month ($20 billed yearly). A Premium seat costs $125 a month ($100 billed yearly), and there's a two-person minimum. Anthropic doesn't train its models on Team account content by default, which matters once staff data is involved.

The Team plan also gives you one bill, one admin page and spending caps for each person. You can mix seat types, too: Standard gets 1.25 times the Pro plan's usage per session, and Premium gets 6.25 times.

Take three office staff, with two on Standard and your heaviest builder on Premium. That comes to $175 a month, or $140 a month billed yearly. It's a sample, not a quote.

The seat is only one factor. In Microsoft's 2026 survey, company support, such as managers backing AI use and real training, was tied to about two-thirds of the payoff people reported. That's how you turn a B into an A.

Who owns what they build for you?

Strong AI users often split their time. You employ them for eight hours a day, and they spend four of those hours on your projects and three on their own.

That's fine. Pay them more than the going rate, as long as you own what they make for you. Put who owns what in writing.

Who owns that work, and how those hours are paid and recorded, depend on federal and state rules. Some states, Minnesota among them, limit what an employer can claim from inventions an employee develops on their own time. Before you set this up, check with your state labor department or an employment attorney.

Even if they stay only three or six months, you've given them the stability they need to improve your business. They'll bring you ideas you never would have thought of.

Common questions

How do I tell if someone really knows how to use AI?

Have them show you: give every candidate the same short task from the real job, let them use AI on it and ask how they checked the result.

What AI interview questions should I ask candidates?

Start with the six questions above, keep them about the work and ask every candidate the same ones.

Should I put AI skills in a job ad?

If the job needs it, say so and explain what it's for. "Drafts customer replies with AI and checks them" beats "AI skills." The post Why your AI-written job ad isn't helping covers the rest.

Fewer people, or 10 more

Say you have three or four strong AI users, and they put real automation in place. Soon you may need only two of them, and then maybe only one.

More likely, though, you'll need 10 of them, because you'll grow that quickly.