Tuesday night, I hosted the first Reflection Series.
The idea behind the Reflection Series is simple: bring thoughtful people with different experiences into a room, introduce a topic worth thinking about, and give ourselves permission to have a conversation without rushing toward an answer.
We’ve built a business culture obsessed with speed. We want faster decisions, faster execution, faster learning, faster answers.
Now AI can give us an answer in seconds.
So what happens when getting the answer is no longer the hard part?
One book that we reflected on was David Epstein’s Range. In it, Epstein explores the difference between what psychologists call “kind” and “wicked” learning environments.
Kind environments have recognizable patterns, established rules, and relatively predictable outcomes. You do something, receive feedback, and adjust. Over time, repetition makes you better.
Chess is a classic example.
Epstein tells the story of László Polgár, who believed exceptional performers could be deliberately developed through intense specialization. He and his wife raised their three daughters with chess at the center of their education. All three became exceptional players. Judit Polgár eventually became one of the world’s best chess players.
It’s an extraordinary story of what specialization can accomplish. But chess is also exactly the kind of problem computers became very good at solving.
The rules don’t change. The board is bounded. Moves can be evaluated, and patterns repeat.
In 1997, IBM’s Deep Blue defeated world chess champion Garry Kasparov.
That doesn’t make the Polgárs’ accomplishments any less remarkable. But it raises an important question about how we’ve structured work.
For decades, we’ve built organizations around many of the same principles that make someone successful in a kind environment. We created departments so people could specialize, workflows so work could be standardized, and certifications to demonstrate mastery of established bodies of knowledge. We created job descriptions asking for increasingly specific experience and built career paths that encouraged people to go deeper and deeper into a particular domain. And we rewarded people for becoming faster and more efficient at executing processes they already understood.
That made sense, but AI is becoming very good at exactly that kind of work.
Wicked environments are the opposite. Epstein borrows a phrase from psychologist Robin Hogarth to describe them: “Martian tennis.” You can see people playing a game, rackets in hand, but nobody has told you the rules. You have to figure them out as you go, and they can change without warning. Feedback is delayed or unclear, and the same move can produce different results depending on timing, context, or who’s involved.
AI can also be useful in wicked environments. It can surface options, identify connections, simulate scenarios, and challenge our first assumptions. But producing an answer is different from exercising judgment.
In a wicked problem, someone still has to decide what matters, whose interests carry weight, what risks are acceptable, and what tradeoffs an organization is willing to make. Those aren’t merely technical questions. They’re human and leadership questions.
What Happens When the Only Problems Left for Us Are Wicked?
Think about the last time a decision at work didn’t have a clean answer. Maybe it was whether to reorganize a team, how to respond to a competitor’s move, or what to do when a longtime process suddenly stopped working and nobody could say exactly why. There was no rulebook to check. The people around the table disagreed about what even counted as success. The problem may involve technology, psychology, culture, economics, and organizational politics all at once, and there isn’t necessarily a best practice because the situation you’re facing may not have existed before.
AI integration is filled with these kinds of problems. Which work should we automate? Where should humans remain involved? How will changing one workflow affect another department?
What happens to trust when someone’s role suddenly changes? How do we capture productivity gains without unintentionally deskilling our workforce? What happens when the technology changes again six months after we redesigned everything around it?
There isn’t a certification that gives you all those answers. And having more information doesn’t necessarily solve the problem. Sometimes we don’t need more data. We need a better frame for understanding the data we already have.
An example in Range that really captured this tension was the Space Shuttle Challenger disaster.
Engineers were trying to understand the risk associated with the shuttle’s O-rings and cold temperatures. On the night before the January 1986 launch, engineers had data from 23 previous flights showing O-ring damage, and the forecast for launch morning was far colder than any of those flights had been. There was data. There were experts. There were analyses. The problem wasn’t simply a lack of information. It was how that information got organized. The engineers presented their concerns to NASA managers using charts that listed each flight in order of launch date, mixing together flights that had no O-ring damage with flights that did, without ever lining them up by temperature. Read that way, the data looked scattered and inconclusive: some cold flights were fine, some warm flights had damage, and no clear pattern jumped out. Had someone plotted the same data with temperature on one axis and O-ring damage on the other, the relationship would have been obvious: every flight below 65 degrees had shown damage, and the coldest flight on record was still far warmer than what was forecast for launch morning. Nobody made that chart. The launch proceeded, and the O-rings failed exactly as the data, properly arranged, would have predicted.
That lesson feels particularly important right now.
AI is making information incredibly abundant. We can ask for another analysis, another scenario, another comparison, another projection, another 20 sources, and we can have them almost instantly. But wicked problems aren’t solved by volume. More information doesn’t necessarily produce better judgment.
AI adds another complication: it can make an answer sound more complete than it is. A well-written response can create the feeling that a problem has been understood when important context, disagreement, uncertainty, or missing evidence has simply been smoothed over.
That makes reflection more important, not less. We need people who can ask:
- What assumptions are embedded in this answer?
- What evidence is missing?
- Who would see this differently?
- What would have to be true for this recommendation to work?
The scarce resource in an AI-enabled organization may not be information. It may be judgment.
There’s early evidence that the work itself is shifting in that direction. PwC’s 2026 AI Jobs Barometer found that the new tasks being added to AI-exposed roles are 2.5 times more likely to depend on skills such as empathy, judgment, and creativity, capabilities that grow more valuable as AI absorbs some routine work.
That doesn’t mean human work becomes easier. It may become less repetitive but more ambiguous. It becomes less about producing an answer and more about evaluating one, less about following a process and more about deciding when the process no longer fits.
AI doesn’t just risk replacing tasks. It’s already sitting closer to the decisions themselves than many people realize. In Deloitte’s 2026 Global Human Capital Trends survey of more than 9,000 business and HR leaders, 60% of executives said they now regularly use AI to support their decisions, and Gartner projects that by 2027, half of all business decisions will be augmented or automated by AI agents. If we consistently delegate framing, analysis, evaluation, and decision-making to it, we may weaken the very human capabilities we’ll need when its answer is incomplete, wrong, or inappropriate for the situation.
Which Raises a Problem With How We Develop People
If the problems increasingly left for humans are wicked, are we developing people to solve them?
I’m not sure we are.
The scale of the challenge is significant. The World Economic Forum reports that employers expect 39% of workers’ core skills to change by 2030. Analytical thinking remains the most sought-after core skill, identified as essential by seven in ten employers.
Yet most development systems still lean heavily toward immediately applicable, role-specific training. We still organize people into silos where they encounter similar problems, similar perspectives, and similar ways of thinking. We still tend to hire the person whose résumé most closely matches the job description. Twenty years in the same industry can look safer than someone who has moved between industries, functions, or careers.
We still tend to view learning through the lens of immediate applicability. What certification does someone need? What skill will make them more productive in their current role? What training produces the fastest return?
And we often fill people’s calendars so completely that curiosity becomes something they have to pursue on their own time.
But range develops differently. It comes from exposure: working with people who see the world differently, learning things that don’t appear immediately useful, moving between disciplines, and trying things you’re not already good at. It comes from encountering enough different situations that, when you face something you’ve never seen before, you have more places to look for an analogy.
This can’t be left entirely to individual initiative. If curiosity, reflection, and cross-functional learning happen only after hours, they’ll remain privileges for people with the most discretionary time and energy.
Organizations that want better judgment may need to design for it: protected time for learning, cross-functional problem-solving, rotations and stretch assignments, structured dissent, and conversations where people can raise questions before a decision hardens into a conclusion.
Maybe Efficiency Isn’t the Goal for Everything
This is why I launched the Reflection Series.
We didn’t gather to become more efficient. There was no workflow to optimize and no certification at the end. Instead, we created space for people with different experiences to think together.
In most organizations, that kind of time can look unproductive, and I think we’re going to have to reconsider that.
If AI saves us time, the temptation will be to fill that time with more work. We produce more, respond faster, make more decisions, and move on to the next thing. That temptation is already becoming real. In a 2026 U.S. Census Bureau survey, 55% of U.S. workers said they’d used AI on the job. Among people who’d used it recently, nearly one-third said it saved one to two hours, while 30% said it saved them at least three hours.
The leadership question isn’t simply whether AI creates capacity. It’s what we choose to do with that capacity.
Perhaps some of that time should be protected to increase and diversify our range.
Read something outside your field.
Talk with someone from another department.
Explore an idea without knowing whether it’ll lead anywhere.
Ask why something has always been done this way.
Sit with a problem a little longer before asking AI to solve it.
Organizations say they want adaptable, innovative thinkers. But we can’t design every minute of someone’s work around efficiency and then wonder why they struggle when confronted with something that doesn’t have an established process.
We May Need to Rethink Expertise
None of this means expertise is becoming less valuable. I think the opposite may be true, but the kind of expertise that matters may be changing.
Deep knowledge will still matter. So will experience. The difference is that we may increasingly need people who can combine that depth with breadth: people who can move between perspectives, recognize patterns across disciplines, question assumptions, and remain comfortable when the answer isn’t immediately obvious.
That should influence how we hire, how we develop people, how we structure teams, how much movement we encourage across functions, and even how we think about someone’s seemingly nonlinear career.
We spent decades preparing people to become exceptionally good at solving kind problems. AI is becoming exceptionally good at many of those same problems.
The next phase of workforce readiness isn’t simply teaching people to use AI faster. It’s developing people who can frame ambiguous problems, test assumptions, integrate competing perspectives, and take responsibility for difficult tradeoffs.
That may require something our efficiency-obsessed workplaces have slowly eliminated.
Time to think.
If this raises more questions than it answers for you and your organization, that’s the point. Start with one: where are you still training people for kind problems that have already become wicked? Sit with that question before you hand it to AI.
Oh, and one more thing…
If you’re in the Atlanta area and interested in joining a Reflection Series, click here for more information.