There's a call that you didn't make this year. You had the read and it was right. And you kept it because you couldn't prove it in that room. You do that a few times a quarter and it lands on no record anywhere. In the next 20 minutes, you will know how to make it safe to be wrong.
Hey there, senior leader, and welcome to the Talent Sherpa Podcast. This is the podcast that helps HR executives move beyond the support function and become business-first, the right hand that your CEO cannot run the company without. Because as you know, the top chair is not an HR job, it is a brand new identity.
I'm your host, Jackson Lynch, and this is episode 161. And it's a Monday, so you have just me. Now, let's dive into this thing.
You have a read on the agent work in your company that you probably haven't said out loud yet, and you've noticed that the people making the call don't have a better read either. That's who this is for.
So here's what we're going to get into today. Why the discipline that made you credible is the reason the biggest calls stop coming to you. How to tell in about five seconds which decisions actually need proof and which ones never really did. And what a year of handing back the ones that you can't prove is costing you today.
Okay, so let's start with a state of play. Imagine this: you're in a room where a call is getting made about the work an agent is now doing, and you have a read on it. You've had it for about 90 seconds, and you know what would have to be true for this thing to run without a person in the loop. And you could say it right now, and you don't say it. And you know exactly why. Somebody would ask how you know. And the honest answer is that you don't. Not in the way that you need to in that room.
So your head of engineering says something less considered than the read that you were holding, and it carries because they're allowed to be wrong about this, and you feel like you're not.
That's the state you're in. In the room, holding the better read, saying nothing, because the only calls that you've ever been able to make are the ones that have perfect provability afterwards.
In Harvard Business Review's 2025 survey of 371 HR business leaders, 91% said that having the right talent is essential to AI success. 21% said HR leadership is closely involved in the decisions about AI strategy. That's the exact same survey with the exact same people. And that one was sponsored by a talent software company, so you need to weigh that accordingly. Nine in 10, though, say that the thing turns on people. Two in 10 have the people executive in the decision consistently. And every quarter that that gap holds, the building gets more practice at deciding this without you.
Okay, it's time for an oxygen check, because when you are climbing at altitude, the very first thing that you lose is your judgment, because you're running out of air and you don't realize it until it's too late. That's why we're here.
So here's the question: when was the last time that we made a call that we could not have defended had anyone asked us to? Take the whole year. Most of us come back with nothing. And nothing is the finding.
We were taught that we only own what we can defend. Now, to be fair, nobody ever wrote that down or said it out loud. It was just quietly true in every room that we've ever had to be right in for long enough that it stopped feeling like a rule and started feeling like a judgment. So we hand back the calls that we can't prove. That's the cost, and it is most of what is arriving right now.
And look, you can't out-execute the way you're thinking about this, which means the fix is not more effort, it's getting your head right first. And that's what we're going to talk about today. So stay with me.
We're going to get into the challenge here. There are two forces that I think are holding this in place. One is ours, and we built it. And the other one belongs to the business, and nobody in it has ever said it out loud. So let's take it in order.
Look at what your function actually produces: a job description, a calibration, a termination file, a comp band. Every one of those things is an object built to survive somebody asking, how do you know? That's how the work is constructed. Every artifact that your function ships, honestly, it has a defensibility step inside of it. The ones without it get taken apart in a room years before you ever arrived.
Now, watch what happens when somebody on your team brings you a recommendation with nothing behind it. You don't say no. You ask what's behind it. And that question is the process, and you didn't invent it. So what I'm trying to say here is the muscle is real and it's the right muscle in most cases, but it only has one setting.
Now, the second one, and this one isn't yours. Your CFO publishes a forecast and comes in under it. Nobody in that building asks whether finance should exist. But they were wrong. Run the same thing on your chair. Be wrong about something once, maybe even about a person in public, and it lands as evidence about the function.
Go look at the last three times a call went badly in your company and ask who got a variance conversation and who got a competence conversation. The right to be wrong is handed out by function, and yours has the least of it. And no one decided that either, in my experience. It just accumulated one incident at a time, and every incident was handled very reasonably in the moment.
So both of those come back to the same thing, I think, don't they?
Here's the reframe. One rule sits underneath both of them, and it has nothing to do with AI agents. You own what you can defend. Nobody ever handed that to you in writing. You built it out of 20 years of rooms where the person with the file kept the room and the person with a read lost it. And it's a pretty good rule. And it's most of the reason people believe you when you do speak, but it only has one setting.
And here's what that setting is costing you right now. Anything that you can't prove stops being yours. And you're the only one who sets it down.
Now you're already probably thinking this is judgment, not caution. Sometimes it is. Hold on to that one. But I'm going to give you one of mine, because I'm working the same problem and I haven't solved it either.
I was helping a client get ready for a board meeting, and we built reads on the people around that table so that she could argue from where they actually sit. So I had the AI agent research them, then ask her about her own history with each one, and then handed back a set of profiles, personas. They were, they were ironclad, they were completely certain. And I couldn't check a word of it, neither could she. She was the source.
We used it anyway. And I think that was the right call, because being wrong there cost a slightly worse rehearsal. And she'd have found out about that live in the room with time to adjust, real time, right?
Now move the same document one room over, same certainty, same nobody can check it, except now it's deciding who sits on what committee.
The bar is not how right it has to be. It's how fast you'd find out that the call was wrong and how quickly you can put it back. And that is, I think, what makes being wrong survivable. You find out sooner and you can undo it. And a mistake you can undo inside a week never gets to become evidence in a court case against you. That's the protection that I don't think we've ever felt like you had. And it was always available to us.
Run that across your own function. And in my experience, most of what you defend comes back reversible. It's a two-way door. You gotta worry about the one-way doors. This is a two-way door in most cases, which means that the bar that you set on purpose lands lower than the one that you inherited more often than it lands higher. And that might be the part that'll surprise you. And it's how you'll know that this isn't a risk exercise. A risk exercise only ever adds.
So take one thing that you review every week, not the big one, just the small one that no one's looked at in the last three years. Ask what actually happens if it goes out wrong on a Tuesday and no one catches it until Thursday. For most of what your function checks, I'd expect the honest answer to be Thursday's fine.
And look, you can read on your own without ever asking anybody. Just next time something goes out and you know it's wrong, watch how long it takes for anyone to notice. And whether the fix is a phone call or, yeah, I don't know. Nobody ever sends a note about that. The lag is the whole reading.
And to be fair, here's where I run out of road on this. I think this holds where a decision has a cost you can name, and it comes apart at scale. One resume that's screened wrong, it costs you nothing. 40,000 of them screens out a whole kind of person. You never met them, so you'll never find out. There's nothing to put back. And I don't really have an answer for that one, other than you don't want to be the talent acquisition software company that gets sued for inherent discrimination. So, like, figure out what we need to do about that.
But what I do know is whose question this is. Your head of engineering is going to tell you the model scores well on the evaluation set and mean it and probably be right. Then ask what happens to the ones it gets wrong. And watch the room work out that nobody actually owns that question.
How sure does this call have to be? And how much proof is that worth buying?
You know, no engineer is going to answer that, because from where they sit, it doesn't even look like a question. Think about what I mentioned about the finance forecast earlier. It is always wrong, but there's no cost associated with it. And yet you've been answering this your whole career. A termination needs documentation, promotion needs calibration. You've been pricing evidence against consequence since you were but a wee generalist, and no one ever told you that it was just one skill at one level.
And there's a discipline for the other half of this, too. And it's been in your plants for roughly a century, and it almost never gets reported to you. Industrial engineers help design work. Sequence, variance, what a step has to produce, where it breaks under load. Like nobody ever called that a people question. But look, in the moment we're in, that's the work of a human systems architect. We haven't had it before, but we've had the skills somewhere else in our organization. We need to adopt that and pull it in. Deciding where the work lives, where the decisions live, and what part of the talent portfolio, whether it be human or agentic or elastic, like all that has to be pulled in together. That's what this system has to prove before it's allowed to act on its own. And in my view, that authority was never assigned. It comes with a chair. So it's ours to grab.
Okay. Here we have the Climb. These are three moves starting on Monday.
Move one, in the next meeting where an agent output is on the table, ask one thing. If this is wrong, what does it cost us? And when would we find out? Five seconds to answer, and anybody in the room can answer it. Where nobody can say when, you found a decision that probably needs you most.
Move number two, take that one and then name the signal out loud. What's the first thing that will tell us that this is going wrong? And how long before it reaches us? Then put a date on when you're going to go look and come back. Now, you've made a call that you can't prove in public with a check on it, and that check, by the way, is what makes it survivable.
Move three, take the signal into the next workflow before anyone can build it. So that's the bar, and the way you should find out is to go into the specification. And the people who design work in your operations group go into that room. If you're not in that room, ask and bring move two as the reason.
Now, none of this has a deadline on it, and that's the only reason it's rarely done.
The discipline that made you credible is the reason the big calls stopped coming. Because it says an unprovable call isn't yours, and every call worth having right now is unprovable. You can tell which ones need proof by how fast you'd find out that you were wrong and how quickly you would be able to put it back. Most of what you check is going to come back as reversible, a two-way door. And a year of handing them back is a year of the building learning to route them someplace else, which will go on and go on and go on doing well after you've changed your mind.
So here's the key takeaway I want you to walk from this episode with. You have been treating provable and yours as the same word. Make the call you've been slowest to notice going wrong. Buckle up and make it anyway.
So before you go, let me thank you for the time. Shout out this week to Curtis from Denver, Colorado. And to everyone who is listening, whether you're in Hamburg, Germany, or Jersey City, New Jersey. The attention that you have is the scarcest thing that you own. And I just want to say thank you for spending some of that with me.
New episodes drop Monday and Thursday, 7 a.m. Eastern. Monday is a solo episode. Thursday is a real conversation with my friend Scott Morris, sometimes even a guest.
And one last thing. Most CHROs are operating without a written mandate, and they don't find out until it's too late to fix quietly. You find out when the calls that you would have gotten right stop arriving. So take 10 minutes, check yours. It's free. The assessment tells you which one of the conditions, there are three of them, that you're probably missing, and the one that's costing you the most. Like I said, it's free. You can find it at mytalentsherpa.com/clarity.
And that's the summit for today. Thank you for climbing with me. And until next time, keep raising the bar. Keep asking what it costs when you're wrong. And keep on climbing!
