Why Speed Is Warping Hiring and Performance

Isabella

Quick note before we start, this episode of Team Pulse is made with AI, including our voices. Everything we discuss is based on real, cited research. So, Jesper, I was looking through this paper published in Management Science from July 15, 2026. It is by Eric VanEpps from Vanderbilt and Einav Hart from George Mason, and HBR just featured it.

Jesper

Oh yeah? What were VanEpps and Hart digging into?

Isabella

They analyzed over 11 million transactions on the freelance marketplace Fiverr, plus a whole series of controlled laboratory experiments. And get this, even a one hour delay in reply time sharply reduced a candidate's chances of getting hired. Like, it actually outweighed higher ratings or much stronger actual qualifications.

Jesper

A single hour? Wow. That is, uh, that is wild because eleven million transactions is not a small sample size. That means decision makers are taking this tiny proxy, response speed, and unconsciously translating it into competence and warmth and future reliability, right?

Isabella

Exactly! They implicitly treat speed as a sign of how good you are at your job. But as the authors point out, speed in an inbox rarely correlates with actual judgment or depth of skill. It just means you are attached to your notifications 24/7.

Jesper

Yes! And in Denmark, you know, we talk so much about mavefornemmelsen, that gut feeling in hiring. Lederweb had that article titled Derfor foeles det rigtigt at ansaette den forkerte, why hiring the wrong person feels right. In our informal culture, god kemi and fast, friendly exchanges mask massive structural biases. We mistake a quick text back for professional excellence.

Isabella

Right. So before you open applications for your next hire or select a vendor, write down in advance whether response speed actually predicts success in that specific role. If it does not, do not let the inbox vote for you. That is rule number one.

Jesper

Yes, exactly. Set the criteria before the emails start rolling in.

Isabella

That focus on speed over quality brings us right into performance management. HBR had another piece in July 2026 called Performance Management Needs New Metrics in the AI Era. And it describes this massive performance paradox happening inside companies right now.

Jesper

A performance paradox? Meaning our existing key performance indicators are completely broken?

Isabella

Pretty much! Organizations are still evaluating workers using pre AI metrics like speed and raw output volume. So the employee who pushes out tons of unchecked AI output looks like a superstar on paper, while the employee who slows down to verify facts, double check logic, and fix hallucinations looks unproductive.

Jesper

Oh, that is so frustrating, but it makes total sense. There was an Organization Science field experiment with 750 knowledge workers using GPT 4. When tasks were inside the AI's capability, people were 25 percent faster and 12 percent more likely to succeed. But on tasks just outside the AI's capability? AI users were 19 percent less likely to reach a correct solution than colleagues who did not use AI at all.

Isabella

Nineteen percent less likely! That gap is where the damage happens. Speed and correctness completely diverge right at the boundary of AI competence, and traditional metrics cannot see that boundary at all.

Jesper

Yes, exactly. Which is why the article proposes three measurement layers. You measure the human contribution, you measure the AI system performance, and then you evaluate them together through what they call a complementarity index. Did the human actually add value beyond the machine, like catching a flaw or reframing a question?

Isabella

Now, HR analyst Brian Heger raised a fair caveat about that. He pointed out that while the complementarity index is great conceptually, operationalizing it is tough. Like, who decides what counts as incremental value, and by what standard?

Jesper

And that is precisely where Danish trust based leadership, tillidsbaseret ledelse, has a huge advantage. You do not solve this with automated tracking dashboards on people's laptops. You solve it in close, regular one to ones. In Denmark, where dialogue is central, leaders can sit down and ask one simple question: when did you last catch an error, or reframe a problem, that the AI did not see? That answer is your true human value.

Isabella

I love that question. Add that to your next one to one agenda. It flips the reward structure immediately.

Isabella

Speaking of overload, back in Week 26 we talked about that MIT Sloan warning, remember? Employees can absorb maybe one or two major changes a year, but leaders keep planning three or four. Well, McKinsey published a piece in July 2026 called Designing for the cognitive cost of change, drawing on nine separate studies from journals like Organization Science and the Journal of Management Studies.

Jesper

Right, treating cognitive capacity, actual mental bandwidth, as a strategic organizational resource. You have to budget it just like capital expenditure or headcount.

Isabella

Yes! And Brian Heger calls the main culprit here goal creep. It is not usually one massive restructure that breaks people, it is the quiet, cumulative load of ten small initiatives added throughout the quarter without removing anything off their plates.

Jesper

It is classic goal creep. Brian Heger actually created some great free one pagers for mapping cumulative change load and doing a mid year recalibration. And, you know, in Denmark, Dennis Noermark wrote a whole book on pseudoarbejde, pseudo work. We have so many tasks running on autopilot that add zero value. Our Danish trust culture makes it easier to speak up and say, hey, this project is useless, let us kill it. But the leader has to take the first step.

Isabella

Right, leaders love launching new things, but nobody wants to host the funeral for dead projects. So McKinsey suggests instituting a strict stop meeting. Before any new project gets approved, the team must explicitly write down what work stops to make room for it.

Jesper

Yes! No written stop list, no new project. That rule alone would save hundreds of hours of burnout.

Isabella

Now, let us turn to what happens long term when we automate all the routine tasks. Back in Week 27, we covered how AI is hollowing out entry level roles. McKinsey Quarterly had an article on July 14, 2026, by Bryan Hancock and Charlotte Seiler, looking at the labor market data. US college grads hit 5.7 percent unemployment in early 2026, and four in ten were underemployed according to NY Fed data.

Jesper

And the Stanford Digital Economy Lab found that young workers aged 22 to 25 in the most AI exposed occupations saw a 16 percent relative decline in employment. But, to be completely honest and fair here, economists are still debating how much of that is AI versus remote work making it harder to onboard juniors at a distance. Yale's Budget Lab says there is no clear economy wide AI fingerprint yet.

Isabella

That is a crucial nuance. But regardless of the exact cause, two senior Microsoft engineering leaders, Mark Russinovich and Scott Hanselman, wrote a fascinating paper in Communications of the ACM in April 2026. They said AI gives senior engineers an AI boost, making them super productive, but gives junior engineers an AI drag, because juniors do not yet have the context to know when the AI is hallucinating.

Jesper

So companies fall into the trap of hiring only seniors and automating juniors, which destroys their own talent pipeline five years down the road! Russinovich and Hanselman suggested a preceptor model, borrowed straight from medical training, where a senior engineer formally mentors a small group, watching how they evaluate AI output.

Isabella

And the practical tool for that is the answer key model. The junior attempts a task completely independently first. Then the AI generates its answer, or acts as the grader, and then a manager sits down to talk through the gap between the junior's attempt and the AI's output. Over time, watching that gap narrow becomes a measurable metric for judgment formation.

Jesper

That medical evidence behind this is incredible! Giving doctors an LLM barely improved their long term diagnostic skills. But forcing them to reconcile their own manual reasoning against the AI lifted their future unassisted performance right up to the AI level. In a October 2025 study in the Journal of Surgical Education, first year medical students using a five day attempt then feedback loop outperformed second year students by a full year of training!

Isabella

A full year ahead in five days! That is astonishing. And companies are acting on this. Bank of America kept its 2026 intake steady at nearly 4000 interns and recruits across 500 plus schools, using heavy simulation to compress judgment building. Their head of global talent, Josh Bronstein, said they have to build those experiences artificially now because routine work is gone.

Jesper

Yes, and that hits close to home here. In Denmark, our entire training philosophy is built on mesterlaere and sidemandsoplaering, practical learning by doing next to a master. As Vaeksthus for Ledelse highlights in their report Pas paa trinet, judgment is built by doing. When AI takes away the basic tasks, osmosis stops working. You have to design the apprenticeship deliberately.

Isabella

So here is your actionable challenge for this month: run the answer key loop with one junior on your team. Let them attempt a task first, compare it with the AI output, and discuss the gap together. Track whether that gap shrinks over time.

Jesper

Yes! That is your new learning metric right there.

Isabella

To wrap up today, ask yourself this final question: where in your organization is fast currently being mistaken for good, and who is paying the price for it?

Jesper

Good chatting, Isabella. See you next week.