AI Can’t Replace HR, But It Can Make It More Efficient

Every few months, I have some version of this conversation with a founder or COO.

“We were going to bring in an HR manager, but honestly, I’ve been using ChatGPT for the handbook stuff, and it’s been fine.”

Here’s the part they don’t expect me to say: It probably has been fine. Truth be told, I’m in ChatGPT a dozen times a day. I’ve used it to draft policy language, build interview guides, write offer letters, translate a statute into plain English, and turn a pile of exit interview notes into something a leadership team can actually read. It has made me faster and more knowledgeable, and I’d tell any small company to use it for all of that.

So, no, this isn’t a post about why AI is the enemy or overrated. It’s a post about the specific thing it can’t do, which happens to be the thing you were hiring an HR manager for in the first place.

Let’s start with what it’s genuinely good at

If you have no HR function and a limited budget, put AI to work immediately on the following:

• First drafts of policies, handbook language, and employee communications
• Job descriptions, interview guides, and structured interview questions
• Offer letters, employment verification letters, benefits explanations
• Plain-English summaries of a regulation you’re trying to understand
• Synthesizing messy qualitative information, including employee survey comments, exit interviews, and manager feedback, into themes you can act on

That list represents a real number of manual human hours that aren’t tied to real value. Refusing to use AI for them isn’t caution; it’s just slower.

What I want you to notice is that every item on that list starts with you already knowing what you need.

The question you didn’t know to ask

AI answers the questions you give it, and it’s not going to tell you that you’ve asked the wrong one. When I bring something to AI, I’m usually about 80% sure of the answer already. For example, if I’m entering a prompt asking about a certain law in a given client’s case, I know roughly where the law sits, I know what I want to say to the client, and I’m using it to pressure-test my thinking and package the result. 

Someone with no HR background isn’t at 80%, they’re at 0%, and the difficult part is that 0% feels the same as 80% when the AI answer comes back polished and confident.

Here’s what that looks like in practice. 

An employee asks to change their schedule. If you ask AI how to handle a schedule change request, you will get a clear, sensible answer about schedule change requests.

The person who has done this for 20 years hears something different. They hear a request that might be an accommodation request in disguise, which starts a process with its own obligations. They wonder whether there’s a health situation underneath it that could trigger leave entitlements. They think about the employee you said no to six months ago and whether saying yes now creates a consistency problem. They note that this conversation needs to be documented today, not reconstructed a year from now when someone’s attorney asks about it.

None of that was in the prompt. AI can’t raise an issue you didn’t mention, and it has no way of knowing what you left out. That’s the gap. It’s the difference between answering a question and recognizing a situation. 

The last 20% is where the liability lives

In my consulting work, the last 20% is judgment and polish. Inside a company, the last 20% is what could show up in a lawsuit.

Think about termination decisions, accommodation conversations, and the back-and-forth they require. Those are human jobs, and they need a trained HR professional to do them. The same goes for workplace investigations, overlapping leave situations, personnel classification decisions, or anything involving a complaint about a manager. 

For all of these, AI can help you prepare. It can outline the process, draft the documentation, and help you think through your questions. It cannot make the call, and the call is the risk.

Four things AI structurally cannot do

It isn’t in the room. It has no institutional memory. It doesn’t know which manager everyone quietly works around, or which two people have a history, or that this is the third time this has come up.

Nobody confides in it. An employee who is worried about something will walk into an office and close the door, not type it into a chatbot. That means the problems you most need to hear about early are left to simmer, and you find out only once they’ve gotten worse.

It can’t be accountable. It can’t sign off on a decision, sit for a deposition, or own the outcome. It would be unclear who is responsible if anything is in dispute.

It can’t say no to you. A good HR advisor will speak up and tell a CEO that what they want to do is a bad idea if they think that’s the case. AI will help you do it faster.

What a small company should do

Stop thinking about it as AI instead of HR, and start thinking about it as three layers.

Drafting and research. Use AI. Often. This is most of your day-to-day volume and it doesn’t need a person except to put in proper prompts. 

Escalation. Write down the short list of situations where someone with real expertise gets involved before a decision is made, like terminations, accommodation requests, investigations, and so on. Having that trigger list written down makes it clear that these are human decisions and require nuances beyond a simple AI prompt. 

Judgment. This needs a human with experience, and it does not need to be a full-time hire. Fractional, outsourced, and on-call HR support exists precisely for companies at this stage, and the cost sits well below the salary you were weighing. 

You’re not choosing between a tool and a headcount. You’re deciding how much expertise you need and how often.

If you’re the HR person reading this

You may have been sent this by someone. Here’s my honest read. 

The risk to your role was never that AI writes a better policy than you do. It’s being the department that’s permanently behind on documentation and therefore never in the room when strategy gets decided. Hand the manual, low-value tasks to AI: the drafts, the letters, the summaries, the first pass at organizing what the employee survey actually said. Then spend the hours you get back on the work that requires being a person in a building full of people.

That’s exactly how I use it, and it’s why I can run the kind of practice I want to run without a large team behind me.