Before You Hire, Ask Yourself the Question You Won't Answer for Your Team

You're not hiring a person. You're hiring an escape.

You feel the bottleneck before you can name it. Something is late, again. You're the one unsticking it, again. And somewhere in the back of your head a thought forms that feels like strategy but is actually panic: if I hire the right person, this stops being my problem.

That thought is the whole failure mode. You're not defining a role. You're casting for a rescuer.

I've watched founders write job descriptions that are really just wish lists for someone to take the pain away without them having to say precisely what the pain is. Urgency does that. It makes you overestimate fit, because you need the answer to be yes. You'll read a resume that half matches and talk yourself into the other half, because you're not hiring for a decision, you're hiring for relief.

The job posting you're about to write isn't a role. It's a description of the part of your job you haven't figured out how to think about yet.

Nobody arrives and takes the whole thing off your hands

Here's the part that stings: the piece of the job you can't describe is the piece that comes back to you. Every time. You hire the generalist, the "just figure it out" person, and eighteen months later you're still the one making the exact same call, because you never actually named what the call was.

This is why "find me someone who can just handle this" is not a hiring strategy. It's a prayer.

Gary, and the standard I had backwards

Thirty years ago, running software teams as a young architect, I interviewed the way I judged myself: hard technical questions, expecting the answer off the top of the head. That's what I could do, so that's what I hired for.

Gary was older, and when I hit him with one of those questions he didn't flinch and he didn't fake it. He said: young man, I don't know the answer to that question off the top of my head. However, I know that the answer is in that book on your bookshelf right now. So when that question comes up, I'm just going to go pull that book down and look it up.

I hired him. We became friends. And the standard I'd been screening for turned out to be wrong.

I wasn't short on people who could recall things. I was short on people who knew the route to an answer they didn't have yet. If I'd kept grading for recall, I never would have found the thing I actually needed, which wasn't a walking encyclopedia. It was judgment.

That's the bottleneck most people never name before they hire. They think they need someone who knows. They actually need someone who knows how to find out, and can be trusted to do it without asking you first.

Delegate thinking, not tasks

This is the whole philosophy underneath everything I teach on delegation: you don't hand someone a task, you hand them a way of seeing. The team isn't failing for lack of a checklist. They're failing because they can't see what you see.

Which is why I don't answer questions. Not rudeness, a refusal to be the human search engine. When someone asks "how should I handle this," the move is "how do you think it should be handled." They articulate their own reasoning, they see the answer themselves, and they grow more in five minutes than a one line directive would ever give them.

Now point that same move at yourself, before you hire anyone.

Use AI the way you'd use yourself, not the way a candidate used it on you

I run every applicant through the same recorded video questions before a live conversation, so the team can hear the person before they meet them. The batch that broke this for me came back nearly identical. Same phrasing, same structure, same rhythm. One candidate had literally pasted the AI's closing line at the bottom of their written answer, the "let me know if you'd like me to draft follow up responses" tag, still attached.

They hadn't answered the question. They'd asked AI to answer it, and forwarded the reply.

That's the exact trap waiting for you when you sit down to write the job post. Open AI and ask it to write you a job description for an operations manager, and you get the same hollow, interchangeable output those candidates sent me. Generic role, generic requirements, generic person. You'll hire someone who fits it perfectly, because nothing specific was ever required to fit it.

The right use of AI here isn't generation. It's interrogation. Before you write a single line of the posting, sit with it and answer, out loud or in writing, the questions you'd force a direct report to answer instead of handing them the fix:

• What decision do I keep making that nobody else is trusted to make?
• What would I need to see in someone's reasoning, not their resume, to hand that decision over?
• If I disappeared for two weeks, which specific call would stall, and why that one?

These three questions look clean on the page. That's an illusion. Sitting with them honestly, instead of flinching back to the job board, is the part that doesn't look clean at all. Most people read that list, nod, and go post the job anyway, because answering it for real means admitting you've been avoiding the answer.

Use AI to push back on your own answer the way Gary pushed back on my question. Not to hand you the role. To make you finish the thinking you were about to skip.

Bottleneck: not the task that's late, the decision you keep making yourself because you've never described, even to yourself, what deciding it well actually requires.

Why this is harder than it sounds

None of this is complicated. It's uncomfortable, which is a different thing.

The fear underneath it: admitting you can't name your own bottleneck feels like admitting you're bad at your job. It isn't. Most people running something have never had to say the sentence out loud because nobody ever made them. Saying it feels like exposure.

The cost: this takes longer than posting a job today. When you're drowning, "figure out exactly what I need first" sounds like the opposite of help. It's the only thing that stops you from drowning in the same spot next year with a new hire's name attached to it.

The rewiring: you have to stop asking "who can take this off my plate" and start asking "who can think about this the way I think about it, on their own." That's a different question, and it doesn't feel as good to ask, because it doesn't promise instant relief. It promises a slower, better one. Expect to circle back to the panic version of the question more than once before the real one sticks.

The comparison

Hiring for the panic	Hiring for the bottleneck
Job post written by AI in five minutes, generic and interchangeable	Job post written after you interrogate yourself on the specific recurring decision
Screens for confident, complete answers	Screens for someone who names what they don't know and how they'd find it
You need the answer to be yes, so you talk yourself into a half fit	You know exactly what a fit looks like because you defined the decision first
Candidate's polish is the signal	Candidate's reasoning route is the signal
The gap in your job comes back to you in eighteen months	The gap gets named and handed over on day one

The diagram

This isn't a straight line to a hire. Most people stall in the middle, loop back to panic, and have to come around again before it sticks.

flowchart TD
    A[You feel the bottleneck] --> B{Urgency: I need this solved now}
    B -- Yes, immediately --> C[You write a rescuer job post]
    C --> Z[Bottleneck survives, resurfaces in 18 months]
    Z --> A
    B -- Resist the urge --> D[Sit with the discomfort: name the recurring decision]
    D --> E{Can you finish the sentence honestly?}
    E -- Not yet --> D
    E -- Yes --> F[Use AI to argue with your own answer]
    F --> G[Write the role around the named decision]
    G --> H[Screen for reasoning route, not recall]
    H --> I[Hire into a defined gap]
    I -.a new bottleneck forms eventually.-> A

Rank this if you only remember one thing

The Gary story is the whole lesson. Everything else, the AI screening trick, the interrogation questions, is downstream of it. If you take nothing else: stop grading candidates, and stop grading the role itself, on whether it produces an instant answer. Grade on whether it shows you the route.

The minor move, useful but secondary, is the AI mechanics. Use it to interrogate yourself before you hire, never to generate the posting for you. If you're screening candidates with any pre-recorded or written response, put a timer on it. Force the question to be seen for the first time when it has to be answered. That's what exposes an AI-fed answer, and it's the same test, really, as asking someone to reason live instead of reciting.

When not to hire at all

If you can't finish the sentence "the recurring decision I need someone else to own is ___" after sitting with it honestly, don't post the job yet. You don't have a hiring problem. You have an undefined one, and hiring into an undefined problem doesn't solve it, it just gives the undefined problem a salary.

Monday

Before you write anything, open a blank document, not a job board, and answer the three questions above about yourself. Then, if you're using AI at all in this process, use it to argue with your own answer, the way Gary would have, the way I try to with every person on my team who asks me instead of themselves.

Write the job post last, not first. If you find yourself dragged back to the panic version of the question, that's not failure, that's the loop doing what it does. Come back around and answer it straight this time.