Two interviews, same quarter. The first candidate had exceptional energy. Fast answers, constant motion, genuine enthusiasm for everything I said. He knew nothing about the problem we were hiring to solve.
The second was incredibly thoughtful. Precise questions, careful reasoning, a real feel for the domain. When we got to the part where you take a first step and learn from what happens, he wanted to understand more first.
I couldn't hire either one. And I've been thinking about why ever since.
The Door Got Narrower
Getting in front of a hiring manager has never been harder. Gartner's survey of job candidates found that 39% used AI somewhere in their application, and more than half of those used it to write the resume. A separate Gartner survey found 6% of candidates admitted to interview fraud: posing as someone else, or having someone pose as them. Gartner predicts that by 2028, one in four candidate profiles worldwide will be fake. That last figure is a forecast, not a measurement, but the direction is not in dispute. The FBI warned in 2022 that deepfakes and stolen identities were showing up in remote job interviews, mostly for software and IT roles.
So the resume is perfect, the cover letter is perfect, and the person may not exist. If you survive all that, you get thirty minutes to make a first impression and, if it lands, another thirty to make a second one with the hiring manager.
Meanwhile the ground moves every month. The tools you'd list on a resume today will be different by spring.
What Survives
When everything written can be generated, the only instrument left is the live conversation. The research agrees: Sackett and colleagues' 2022 re-analysis of the selection literature, published in the Journal of Applied Psychology, found structured interviews to be the top-ranked predictor of job performance.
A live conversation can't be pre-generated. It shows what a person does with a problem they haven't seen. In my experience, two abilities show up there that no tool change touches.
Understand
I've written about rigorous curiosity, and it's still the entry fee. But curiosity alone isn't enough to solve a problem.
You have to care about the people living inside it. You have to absorb a new domain quickly and apply it in context. You have to communicate with people and with machines, and know which one you're talking to. These are teachable, and they take reps.
Then there's the part you can't teach. It can't be rote. You can hear when someone is going through the motions; the answers are correct and have no pull. The people who solve problems want to see the realization land. They want the other person to get where they were trying to go. That's the payoff: not the paycheck, not even the work. The outcome.
The tell is specificity. Enthusiasm can be performed. Detail cannot: what they noticed, what they tried, what they would do first.
Move
Here is why this matters more than it used to. When the cost of production collapses, there is no moat. I made the case in Personal Software Is the Hero Government Software Needs: you can't spend millions to buy yourself time and space anymore. A lead is measured in months, at best.
What's left is how well you understand the problem and how eager you are to help someone solve it. That second half is urgency, and it is widely misread.
Some AI startups now put seventy-hour weeks right in the job listing. The schedule has a name: 996. It treats hours as a proxy for commitment, which is the same mistake as counting tokens. Amazon told its engineers this week that token usage and AI adoption rates are "deeply flawed" measures, and that it is best to measure, and be measured by, long-term outcomes. Effort without a target is heat, not work.
Urgency, properly understood, is one movement with several parts. Find the most valuable thing to move on. That requires understanding the problem and the domain. Then move on it with the person who owns the problem, and let their feedback change your direction. The agile basics were never about speed; they were about shortening the distance between a decision and the evidence about it.
Two Ways to Fail
That brings me back to the two candidates.
The first had energy and no understanding. Motion without direction is expensive, and AI made it cheaper to generate. The second had understanding and no movement. Analysis without feedback is just a more sophisticated way to be wrong later.
You need both, and not in sequence. Understanding tells you where to move. Moving tells you whether you understood.
It is a hard skill, and it sounds simple in conversation. It takes reps, the same as any other. I made that argument in 2025 about skills generally. It applies more to this one than to any other, because you can't fake it for thirty minutes with someone watching.
What I Listen For
I'm not looking for the best resume or the best answers. I'm listening for questions about the problem before questions about the role. For someone who asks what the last person got wrong. For a person who, given a vague situation and a few minutes, picks a first step, explains why it's the right one, and changes it when I push back.
The tools will change by the month. AI can write the resume, answer the question, and now borrow the face. It can't want the outcome for you.