What only a human can do (even when AI could)

The HRBP Lab · Issue 003 · May 16, 2026

Hello from the Lab and welcome! Let’s get into it.

Third issue in a short arc. Issue 001 was about the work we should stop doing entirely — deliberate subtract. Issue 002 was about the work we should hand to AI because it doesn’t need our judgment a second time — deliberate codify. This one is the third decision. The work we keep human on purpose, even when AI could plausibly do it. Deliberate protect.

Links to the first two are at the bottom if you joined recently.

A leader I support called me last week. She had to give her own manager hard feedback and wanted help finding the words.

Forty-five minutes on the call. About five of those were on the actual script. The other forty were her, talking through what was around the conversation — what was at stake, why the trust had eroded, what she needed her manager to hear.

She could have asked Claude to draft the feedback. She didn’t. She came to me.

Here’s what I noticed. There was a category of work happening in those forty-five minutes that AI couldn’t do, and it wasn’t the drafting. The drafting was the easy part. The work was that someone — an actual person who knew her, who had context on the relationship, who could hear the silence between the words — was on the line with her while she figured out what she actually wanted to say.

The output of the call could have been an AI-generated paragraph. It would have been polished. It would have been accurate. It would have been wrong.

The category we’re at risk of losing

There’s a category of HR work where the output looks identical whether a human or an AI does it. A piece of feedback. A note of recognition. A conversation about a struggling teammate. A goodbye to someone leaving the company.

AI can produce the words for any of these. The words aren’t the work.

The work is that a human in the seat — someone with skin in the game, someone the recipient knows is listening — did it. Recognition lands because the person reading it knows you wrote it. Feedback lands because you chose to give it. The 1:1 lands because you showed up. AI doesn’t change this. It just makes it easier to forget, by making the words faster to produce.

I use AI all day. This is not an anti-AI piece. The whole thesis is that AI can do these things, and we’re choosing, on purpose and not by accident, to keep them human. That choice has to be made out loud, or it won’t be made.

Why this is HRBP work

The HRBP role exists, in part, to remind leaders that there are people at the center of every decision the work serves. That isn’t a tagline. It’s a job function. One of the many incredible HR leaders I’ve had the pleasure to work for called HR the “conscience of the organization.”

When a leader asks “should I just have AI draft this?” — the HRBP is often the only person in the room who can say “no, and here’s why.” The instinct to refuse is correct. The next move is turning the instinct into a framework: name the line for the team before they bump into it, not after.

This is the part of the AI conversation that isn’t being had. Most of the conversation is about what to automate. The harder conversation is what to protect, why, and how the team will know the line is real.

The output isn’t a manifesto. It’s a small list. Three or four items. Specific to your function. Defended out loud.

Five questions to find your line

Take these into your next 1:1 with a leader you support. Don’t run all five. Pick two or three:

1. What did you do this week where the value was specifically that you did it?
2. If your team found out AI helped you draft this, would they feel differently about it?
3. Where in your work does being seen by you matter more than the decision being optimal?
4. If you offloaded this kind of decision to AI, what would your team lose, even if the output was identical?
5. What part of your job exists specifically to protect a person from being treated as a number?

Each question points at a candidate “human-only” item. The leader walks out with one clearly-named line: I will personally write this, I will personally show up to this, I will personally make this call. You walk out with the input for naming the rest of the team’s lines collectively.

If you ran the Issue 002 questions, this is the inverse. The two question sets are halves of the same job: what to hand off, and what to protect.

My own version of this

Sunday I look at the upcoming week and ask: what can I codify? (Issue 002.) Then I ask: what stays mine?

This newsletter is on the second list. I draft these by hand. AI helps me think about structure, surface tension in a draft, stress-test where my logic breaks. The words are mine. That choice is on purpose. I’d rather send fewer issues I wrote than more issues someone else’s machine could have produced.

I don’t say this for credibility points. I say it because the choice has to be visible. That’s how the rest of the team knows the line is real.

Before you do anything

Before your next AI rollout, name two or three pieces of work that stay human in your function, and tell your team. Not “we value the human element” abstractly. The specific items, the specific reasons, in language the team can repeat back to you.

What’s the first thing on your human-only list? Hit reply and tell me. I read everything that comes in.

Subtract. Codify. Protect. That’s the deliberate decision set. Now we get to talk about everything else.

Next issue, I’m picking up a thread that’s been running underneath all three: HRBPs as builders. Not just users of AI tools — designers of them. More on that next issue.

-Josh

The prompt I’m using this week

This pairs with the five questions above. Use it in a 1:1 with a leader you support. Same setup as the Issue 002 prompt, inverted from “what’s repeatable” to “what’s irreplaceable.”

I’m helping [LEADER NAME] name the parts of their work that should stay human even when AI could plausibly do them. Help me run a short discovery on the following five questions. For each one, suggest a follow-up I can ask if their answer is short or vague. After the conversation I’ll come back with their answers and we’ll draft a short “human-only list” for their team — three or four items, with the specific reason each one stays human.What did you do this week where the value was specifically that you did it?If your team found out AI helped you draft this, would they feel differently about it?Where in your work does being seen by you matter more than the decision being optimal?If you offloaded this kind of decision to AI, what would your team lose, even if the output was identical?What part of your job exists specifically to protect a person from being treated as a number?

Why it works: The five questions are uncomfortable to sit with alone. The discomfort is the point. The prompt makes Claude do the structural work of pulling answers deeper so you can stay focused on the person in front of you. The “we’ll draft a list together” framing turns the 1:1 into the front end of a real artifact, not the front end of a strategy doc.

I’ve started running the inverse pair (this prompt plus the Issue 002 prompt) over two consecutive weeks with leaders I support. The first 1:1 surfaces the codifiable work. The second draws the line on what stays. Two halves of one move.

If you try it, tag it #buildtogether and show me what you built. I’ll feature the best ones in a future issue.

If you want to go to the source:

• The Extra Time Challenge — Issue 001, 2026-04-25 (the deliberate-subtract setup)
• What did you write this week that you’ve written before? — Issue 002, 2026-05-09 (the deliberate-codify middle move)

If someone you know is navigating the same question, forward this their way. New here? Subscribe at thehrbplab.ai to get the next one.