As AI absorbs more of the routine analysis that once filled a typical workday, the decisions that still land on a person’s desk are precisely the ones that were never going to be routine in the first place. They are ambiguous, they carry real consequences, and they rarely come with a clean precedent to fall back on. That shift is producing what researchers now describe as a judgment gap, a widening distance between how much organizations depend on sound human judgment and how little most people are still given the chance to practice it.
The gap is not a hypothetical risk sitting somewhere in the future. It is already showing up inside organizations that adopted AI quickly and are only now noticing that judgment, once assumed to develop naturally on the job, needs a deliberate place to grow if it is going to keep pace with everything being automated around it.
The Paradox at the Center of Automation
David S. Duncan’s research, published in the Harvard Business Review, identifies a genuine paradox sitting at the heart of most AI adoption. Judgment, defined as the capacity to make wise decisions under uncertainty by weighing quality, context, tradeoffs, and consequences, has traditionally developed through hands-on responsibility and repeated exposure to outcomes. AI increases the organizational need for that exact capacity at the same time it quietly removes the everyday tasks that used to build it, particularly for employees early in their careers who once learned judgment by doing the routine work themselves.
That combination creates a slow-moving risk that is easy to miss in the short term. Judgment ends up concentrated in a shrinking group of senior people who developed it years earlier under different conditions, while the pipeline meant to replace them, gradually, through experience, has far fewer of the formative moments that judgment actually depends on.
Judgment Is Not the Same as Speed
It is tempting to treat a faster decision as a better one, especially when an algorithm can produce a plausible answer in seconds. Research from Martin Reeves, Mihnea Moldoveanu, and Adam Job, published in the Harvard Business Review, pushes back on that assumption directly, arguing that many of the most consequential parts of decision-making sit outside what data and algorithms can reach on their own. Framing the problem correctly, choosing which data sources deserve trust, imagining possibilities nobody has tested yet, and applying a values judgment to a genuinely new situation all remain distinctly human contributions, and the growing sophistication of AI tools appears to be making those contributions more differentiating rather than less.
That distinction matters because it reframes what leaders should actually be protecting. The goal is not to slow down every decision in the name of caution, and it is not to defer automatically to whatever a model recommends either. The goal is preserving the specific human capabilities that no dataset can substitute for, and building organizational habits that keep those capabilities sharp rather than letting them atrophy quietly in the background.
Staying “Upshifted” Instead of Reacting on Autopilot
FutureThink’s Kill the Company workshop programming is built around a critical gap many senior teams share: the need to think critically and act decisively rather than defaulting to whatever recommendation, human or automated, happens to arrive first. The underlying model FutureThink teaches distinguishes between two states of mind, referred to internally as upshifting and downshifting. Upshifting engages the neocortex, the seat of critical and innovative thinking, while downshifting drops a person into the limbic system’s habits and emotional shortcuts or, under real pressure, into the brain stem’s raw survival instincts.
A downshifted team tends to accept the first plausible answer in front of it, whether that answer comes from a colleague, a dashboard, or a generative AI tool, simply because pausing to evaluate it feels like friction. An upshifted team treats that same recommendation as a starting point rather than a conclusion, deliberately re-engaging critical thinking before anything gets acted on. Practicing that shift is less about willpower and more about building a repeatable habit, which is exactly what structured exercises are designed to reinforce.
What Leaders Can Do About It
Duncan’s research points to a specific, practical fix rather than a vague call for more oversight. Organizations need to redesign work deliberately so that judgment still gets built, clarifying who actually owns a given decision, exposing people to the real consequences of their choices instead of shielding them from outcomes, and restoring the kind of stretch assignments that once came bundled with entry-level work. Simulations and case-based learning can recreate some of that experience artificially when the real version has been automated away, but only if organizations treat this as deliberate design rather than something that will simply take care of itself.
Complementary research from MIT Sloan Management Review adds a related point worth building into that redesign: individual decision-making style shapes how much value a person actually gets from identical AI advice, with some executives investing meaningfully more in strategic initiatives based on the exact same recommendation than others. Organizations that understand those differences are better positioned to know where a human’s judgment adds the most value and where it needs the most active protection.
FutureThink’s broader Navigate Change programming and its position that AI automates expertise but cannot replace human judgment both point toward the same conclusion. The organizations that come out ahead will not be the ones that automate the most. They will be the ones that were deliberate about which decisions still needed a person, and that built the structure to keep that person’s judgment sharp.