There is a comforting story people like to tell about technological revolutions. A new tool appears, productivity rises, old inefficiencies disappear, and society, after a little turbulence, moves to a better equilibrium. It is a clean story. It is also, very often, the wrong story.
The AI revolution does not feel clean because it is not arriving in
a neat, predictable world. It is arriving inside markets, firms, labor systems,
schools, bureaucracies, media networks, and geopolitical rivalries that are
already unstable, interdependent, and constantly adapting. In other words, AI
is not landing on a chessboard. It is landing in weather.
This is where Scott Page’s idea of landscapes becomes useful. It gives us a way to think about decisions in complex systems without pretending the world is simpler than it is. The basic metaphor is easy: imagine that every possible strategy, policy, organizational design, or technology choice sits somewhere on a landscape. Higher ground means better performance. The question is how easy it is to climb.
In a simple landscape, the path is fairly obvious. Improve a little, and things usually get better. You can make sensible local decisions and trust that they are moving you uphill.
In a rugged landscape, things get trickier. There are lots of peaks, not just one. You can climb successfully and still end up stuck on a mediocre hill, unable to reach a better one without first going downhill. The danger here is local success. You improve, but into a trap.
Then comes the nastiest version: the dancing landscape. Here, the terrain itself moves while you are climbing. The hill that looked best this quarter may not be the best one next year. The map changes while you are still folding it open.
Once you see these three landscapes, a lot of the AI conversation starts looking different.
The most naive AI story assumes a simple landscape. Build better models, deploy them, automate tasks, cut friction, raise output, everyone adapts, done. That is the glossy presentation deck version. It assumes there is one main hill called “efficiency,” and that all serious people should sprint uphill.
But the real world looks much more like a rugged landscape. Firms are making decisions under uncertainty, and many of those decisions interact. A company cuts headcount because AI tools let the remaining employees do more. Margins improve. The stock likes it. Management concludes it has found a better organizational form. Maybe it has. But maybe it has just climbed the nearest visible hill.
That is the first hard lesson of rugged landscapes: short-term improvement does not prove long-term wisdom.
A leaner company may be faster in routine execution yet weaker in resilience. It may have removed people who looked redundant on a spreadsheet but who carried institutional memory, mentoring capacity, or a different way of thinking. It may have improved throughput while damaging adaptability.
This is
especially relevant because Page’s broader work on complexity and diversity
keeps returning to a stubborn idea: on hard, nonroutine tasks, systems often
benefit from a wider repertoire of perspectives and tools, not just from
squeezing more output out of a narrower elite. That matters in AI because many
executives are currently acting as if “smaller team plus better tools” is
self-evidently the final form of intelligence. It may turn out to be the final
form of quarterly confidence instead.
The labor angle is the clearest place to see this. When people ask whether AI will “replace jobs,” they often picture a dramatic one-to-one substitution: software does task X, therefore worker Y disappears. Sometimes that will happen. But the more important story is structural. In a rugged landscape, the labor market changes through chains of local decisions that reinforce one another.
A firm hires fewer juniors because AI makes seniors more productive. Another firm does the same. Entry-level openings shrink. Training ladders weaken. Fewer people get the chance to build real expertise. Bargaining power shifts toward employers. Workers who remain employed are expected to cover more ground. Productivity may rise, but so does precarity. Eventually politics notices.
That is not a Hollywood robot takeover. It is something quieter and probably more realistic: a slow rewiring of the routes by which people enter, learn, advance, and negotiate. The system changes shape before it announces that it has changed shape.
And then, just to make things more interesting, the landscape starts dancing.
This is where the AI story gets genuinely weird. It is not only that firms are adapting to AI. AI itself is changing at high speed. Model capabilities improve, costs shift, interfaces evolve, regulation stirs, competitors copy one another, public trust rises and falls, and entire categories of work are redefined in real time. The landscape is moving because all the agents on it are moving too.
That means a strategy that looks brilliant today may simply be
overfit to the current moment. A company may reorganize around the belief that
AI will mostly automate execution, only to discover that the real bottleneck
becomes judgment, distribution, trust, or integration. A university may
redesign training around prompt fluency only to find that everyone has prompt
fluency and the scarce skill is now domain depth. A government may regulate
yesterday’s frontier while tomorrow’s frontier walks around the rule.
This is why the “musical chairs” feeling around AI is not imaginary. People sense that they are being asked to adapt before anyone can say, with confidence, what stable adaptation even looks like.
The business implications are huge. In a dancing landscape, over-optimization is dangerous. You do not want to strip away all slack, all redundancy, all experimentation, and all human variety in the name of immediate efficiency if the environment itself is still changing. That is a great way to become extremely efficient at solving last year’s problem.
The geopolitical implications are just as large. AI is not only about chatbots and office software. It touches military planning, cyber operations, intelligence analysis, semiconductor supply chains, infrastructure security, industrial policy, and state capacity. Countries are not choosing whether to adopt one tool; they are repositioning themselves inside a contest whose rules are still evolving. That is classic dancing-landscape territory. Every move changes the payoff structure for the next move.
Even culture starts behaving differently under this framework. AI affects not only what can be produced, but what can be trusted. It changes the economics of attention, the meaning of authorship, the cost of persuasion, and the volume of synthetic noise in the system. When abundance rises but trust falls, the peak is no longer just “more content.” Suddenly the valuable hill may be authenticity, curation, or institutions that still carry signal.
This is why complexity thinking helps. It refuses both the cheerful fantasy that technology automatically lifts all boats and the equally lazy fantasy that everything is doomed. Instead, it asks a better question: what kind of landscape are we actually in?
If the answer is simple, then optimization works well. If the answer is rugged, then exploration matters because local maxima are traps. If the answer is dancing, then resilience, optionality, and adaptability matter even more than efficiency.
That last point has political consequences. In a moving landscape, the job of public policy is not to guess the final destination perfectly and then force everyone toward it. That is planner vanity. The real task is to help society adapt without breaking. That means protecting people from being crushed by transition while preserving enough flexibility for experimentation.
So the right response to AI is probably not a single grand policy with a heroic name. It is more boring than that, which is often how serious things look in daylight. Wage insurance. Portable benefits. Recurrent retraining tied to real demand, not motivational brochures. Stronger worker bargaining power. Competition policy that prevents a few firms from locking up the gains. Education that teaches people how to think across models and domains rather than just training them for one brittle slot in a machine.
In other words: if the terrain is moving, build people and institutions that can move with it.
That may sound less exciting than the usual AI prophecy, but it is
also more honest. The danger in moments like this is not just that people
underestimate the technology. It is that they overestimate their ability to
stand still while the world reorganizes around them. Firms can get trapped on
the wrong hill. Workers can be shoved off the map. States can optimize for a
version of the future that never arrives.
The point of the landscape metaphor is not to make everything sound mystical. It is to inject a little humility into systems that are drunk on certainty. AI may indeed create extraordinary gains. But in a rugged, dancing landscape, gains are not enough. The real question is whether the path that looks smart right now leads somewhere durable, or whether we are all just climbing very confidently toward a hill that is already moving.
