As organizations automate more decisions, what happens to human judgment? An exploration of Human Debt, automation bias, and organizational resilience.
July 9, 2026
We’ve gotten very good at the how of business. Algorithms forecast demand, screen resumes, monitor keystrokes, and optimize headcount down to the decimal point. What we haven’t gotten very good at, and what remains inconveniently human, is the why. And nowhere is that more visible right now than in the way AI hiring bias shapes decisions we assume are neutral. That gap is where the risk lives. It’s also the question I keep thinking about: why are we so much better at automating tasks than at understanding what we lose in the process?
I spend a lot of time trying to connect the dots between evidence, narrative, and what’s actually happening inside organizations. Over the last few years, I’ve noticed a shift in how some leaders talk about their people. Less in terms of growth, more in terms of output, even while the language of “development” stays the same. Duena Blomstrom’s idea of Human Debt is a useful way to think about why that matters. Like technical debt, human debt builds when organizations trade psychological safety, trust, and institutional knowledge for short-term speed and efficiency. I’m using her term deliberately because I think what’s happening now, with AI speeding up those trade-offs, deserves the same name and more attention than it’s getting.
The Algorithmic Trap
The logic is simple: automate the decision, remove the bottleneck, lower the cost. In practice, that usually means fewer recruiters, fewer managers, and fewer humans left in the loop.
But this is where a lot of leadership teams blur something important: efficiency is not resilience. An organization built for calm weather, predictable inputs, and no real pushback anywhere in the system can look strong. A lot of the time, it’s just brittle. And brittle systems do not bend well when something changes. They break.
Algorithms are very good at pattern matching under known conditions. They’re much weaker at the things that usually matter most when a company is under pressure: judgment when the situation changes without warning, and the ability to sense morale slipping before it shows up in a metric. Capability is not wisdom. A system can be excellent at optimizing for the wrong thing. I don’t think most leaders have really sat with that.
AI Hiring Bias and the Human-in-the-Loop Fallacy
Most leaders think they’ve solved this. “We have a human in the loop,” they say, as if the phrase itself settles the question.
It isn’t, and there’s research to back that up. Research from the University of Washington suggests that AI doesn’t just automate decisions. It can shape them. When participants were given strongly biased hiring recommendations, their choices lined up with the AI’s recommendations about 90 percent of the time.
A separate Stanford analysis of four million job applications across 150 employers found something more structural: when multiple employers rely on the same AI screening tools, the same candidates can be screened out again and again across different companies because the systems tend to make similar decisions. In other words, the bias does not stay contained inside one company. It can spread across employers using the same tools.
This is not some fixed law of human nature. It’s mostly a design problem. When systems are built to move fast, with no real time to think and no real authority for a human to push back, people tend to take the path of least resistance. That’s fixable. But most organizations haven’t fixed it yet, not necessarily because they’re careless, but because AI is still new for most of them, mostly used for administrative work, and because the harder question of what real oversight actually looks like still hasn’t been taken seriously.
Talk to enough AI vendors and the recruiters using their tools and you hear the same thing: “human review” of an AI-screened shortlist often takes under ninety seconds per candidate. That raises a harder question for every leader running one of these systems: if the human reviewer isn’t actually empowered to override the machine, what exactly are they there for?
The Accumulation of Human Debt
Every time an organization trades human judgment for an automated shortcut, it doesn’t remove the cost. It just pushes the cost further out. That debt is growing faster now than when Blomstrom first described it.
The interest shows up in pretty predictable ways. Institutional memory erodes as experienced people are automated out or leave. Psychological safety weakens when employees feel monitored instead of trusted. Discretionary effort, the extra people give freely when things go wrong, starts to disappear too. The organization appears efficient in calm conditions but struggles when conditions change. Over time, it can end up pinned against a wall it built itself.
The Great Divergence: Compliance vs. Agency
From where I sit in executive search and talent advisory, I think something is shifting in the way talent is valued. I want to be careful here, because this is not a neat split between two kinds of people. Most roles, and most people, contain both compliance and agency. What’s changing is which of those two capacities organizations are actually paying for.
Roles built mostly around following algorithmic nudges and executing repetitive tasks inside tight parameters are easier to benchmark and automate. That’s not a judgment on the people doing them. It is a statement about how the roles were designed. What I keep seeing is that the people who can question the premise, change course when the situation demands it, and use real judgment are becoming harder to find and more valuable to keep. That is because this capacity is exactly what an algorithm cannot replicate.
I don’t have a clean dataset proving that individual agency creates enterprise value. No one really does, at least not with “agency” treated as a separate, measurable variable. I’d be skeptical of anyone claiming otherwise. What I do have is years of watching which hires actually help hold an organization together when things go sideways. It is rarely the person who followed the process perfectly.
The One Question
I don’t think this is really a framework or checklist problem. I think it comes down to one question every leader should ask honestly and sit with for a while.
Are you willing to bet your organization’s future on no longer needing humans in the room, not because the algorithm got smarter, but because you decided the humans had nothing left to add?
If the honest answer is yes, that’s a real strategy. Build for it deliberately. If the honest answer is no, then every automation decision you’re making right now deserves the same scrutiny you would give a debt you’re not sure you can repay.
Related Insights
Why AI Makes Broken Workplaces Worse, Not Better