Everyone asks what AI will replace. The better question is what becomes newly valuable once intelligence gets cheap.
Every big technology arrives wearing the disguise of a labor-saving device. At first, we describe it by the task it automates. Only later do we see the system it rearranges.
The car looked like a faster horse. It turned out to be a machine for suburbs, motels, drive-throughs, teenage freedom, oil politics, and a completely new geography of everyday life. The first-order effect was transportation. The real story was everything transportation touched.
I think AI is going through the same misunderstanding now.
The public conversation is still dominated by first-order questions: which jobs will AI replace, which tasks will get faster, which firms will cut headcount first. Those are not bad questions. They are just early questions.
The more interesting ones sit a step downstream. What happens once intelligence is cheaper? What becomes newly economical? Where does demand expand instead of contract? What new bottlenecks appear? And what kinds of work grow around those bottlenecks?
That is where the job story gets more interesting.
The economy is not a list of jobs. It is a map of frictions.
Most people think about technology in straight lines. A tool automates a task; the person doing the task becomes less necessary. Sometimes that happens. But it misses how systems actually behave.
The economy is not a neat list of occupations. It is a map of frictions. Remove one friction and the pressure shows up somewhere else.
If AI makes analysis cheap, more people start asking for analysis. If AI makes tutoring available on demand, more people attempt harder learning. If software becomes easier to build, more firms can afford custom tools. If content becomes abundant, the scarce thing is no longer content. It is trust, taste, and selection.
Second-order thinking is really just the discipline of following that pressure.
Interactive
Where Does the Bottleneck Move?
Pick a shift. See what becomes abundant, what becomes scarce, and where new work forms.
New work that forms
Once you trace where the bottleneck moves, AI looks less like a machine that simply destroys jobs and more like one that shifts them. And jobs, more often than not, form around bottlenecks.
When expertise gets cheaper, people use more of it
This is the part I think people underestimate most.
The standard story says that if AI makes expert knowledge easier to access, experts become less valuable. But that only makes sense if demand holds still. In practice, demand rarely holds still when something important gets cheaper.
When something useful becomes easier to get, people use more of it.
The spreadsheet did not make organizations want less analysis. It made analysis cheap enough to use everywhere. Search did not reduce the need for research. It made research a daily reflex.
AI is likely to do something similar for expertise.
More small businesses will seek advice they used to skip. More managers will run scenarios they once guessed through. More consumers will navigate complicated decisions with more confidence. More students will attempt subjects they would have written off. More people will try to start things because the cost of getting competent help at the beginning has collapsed.
That does not eliminate experts. It changes where experts sit in the value chain.
The premium shifts toward people who can do the last mile: adapt general intelligence to a specific situation, make judgment calls when the facts are incomplete, and take responsibility when the stakes are real. That means more value for implementation specialists, reviewers, auditors, trainers, coaches, translators, and operators who can turn a generic model into a useful decision inside a messy context.
When expertise becomes abundant, applied judgment gets more valuable.
When production gets cheap, trust gets expensive
There is another second-order effect that matters just as much.
AI can make words, images, code, plans, and recommendations abundant. At first glance that looks like a story about supply. But the more important story is what happens to scarcity.
In a world where almost anything can be generated instantly, the scarce resource is not output. It is confidence. Which report is reliable enough to act on? Which synthetic image is authentic? Which automated recommendation is sound? Which summary actually captures the nuance? Which brand still signals a human standard behind the machine?
Every technological wave creates a new scarcity. AI seems likely to create a scarcity of trust.
And scarcity creates work.
Some of that work will look like verification, certification, review, provenance, quality assurance, compliance, editorial judgment, and reputation management. Some of it will look like new businesses devoted to proving that a system is reliable, traceable, fair, or aligned with a customer's needs. Some of it will simply look like a human being willing to put their name on the output.
The digital economy spent years making creation cheap. The AI economy may spend the next decade making confidence legible.
That is not a side effect. It is a labor market.
When customization gets cheap, the market fragments
There is a third effect that matters because it changes who gets to build businesses at all.
AI lowers the cost of customization. If software, services, and workflows can be tailored more cheaply, firms no longer need enormous markets to justify making something specific. Suddenly it becomes economical to build for narrower industries, smaller customer groups, and more idiosyncratic problems.
That means the future may be less standardized than people expect.
Instead of a handful of generic tools serving everyone badly, we may get a long tail of services serving smaller groups well: systems for school districts, property managers, dental labs, regional logistics firms, independent retailers, elder-care providers, contractors, specialty manufacturers.
When production is expensive, the market tends toward sameness. When production gets cheap, difference starts to pay.
That creates new roles for the people who understand those differences: consultants, workflow designers, integration specialists, niche operators, product translators, and founders with real domain knowledge. One of the underappreciated effects of AI may be that it makes local and industry-specific knowledge more valuable, not less.
Generic intelligence is powerful. But somebody still has to teach it where the bodies are buried in a particular business.
When coordination gets easier, smaller firms get bigger
Large companies have always had one quiet advantage: they can absorb coordination. They can manage vendors, handle reporting, standardize processes, keep up with compliance, and carry layers of administrative work that would overwhelm a smaller team. AI chips away at some of that coordination cost.
That means a small firm can behave like a much larger one.
A five-person team can produce better proposals, manage more customer communication, automate more follow-up, handle more internal reporting, and support more clients than it could a few years ago. That changes the structure of competition.
Some work that once lived inside big firms will move outward into networks of smaller firms and specialized providers. A company may not need a full internal department if it can buy a mix of software, fractional expertise, and AI-enabled operations from outside.
Again, that is not the disappearance of work. It is its rebundling.
The roles that grow may not always be glamorous, but they will matter: the people who configure systems, keep data clean, train teams, manage vendors, oversee compliance, and make sure the promised efficiency actually shows up in practice.
AI may do for coordination what cloud computing did for infrastructure: give smaller players capabilities that used to belong to scale. And when smaller players can compete, they hire. Just differently.
Take a future where most cars drive themselves
Even here, the interesting story is not the most obvious one.
If most cars eventually drive themselves, people will rush to debate the fate of drivers. That is understandable. But the more revealing question is what abundant mobility makes newly valuable.
Interactive
Follow the Chain
Pick a first-order effect. Click to unfold the second-order consequences.
Self-driving becomes mainstream. Then what?
Start with time. If travel becomes less demanding, the car stops being dead time and starts behaving more like usable time. That changes what businesses can sell during a trip: entertainment, commerce, work tools, maybe entirely new service formats built around motion rather than destination.
Then think about the service layer. A world of self-driving cars does not run itself. It needs charging, cleaning, maintenance, dispatch, remote support, curb management, software monitoring, fleet logistics, interior design for new use cases, and city planning that treats pickup and drop-off as a core piece of infrastructure.
Then think about land. If parking demand changes, some of the most valuable urban real estate in the country gets re-priced and repurposed. That creates work for planners, developers, regulators, designers, and the businesses that follow newly available space.
The point is not that every old job neatly becomes a new job. It will not. The point is that automation rarely leaves a vacuum. More often, it creates a whole new layer of work around the system that replaces the old one.
We have seen this pattern before. Cheap computing did not simply eliminate clerical labor. It created IT, cybersecurity, SaaS, digital marketing, product management, cloud infrastructure, and entire categories that would have sounded bizarre a few decades earlier.
Cheap mobility will have its own stack. Cheap intelligence will too.
The better question
The most misleading question in AI is: what will it eliminate?
That framing encourages people to look at the existing map of work and imagine certain boxes disappearing. But that is not how big technologies reshape an economy. Once a capability becomes cheaper, people reorganize around it. They raise expectations. They attempt things that were previously not worth attempting. They build businesses that were previously not economical. And in the process, they create new bottlenecks.
That is where the new work comes from.
So the better questions are what becomes abundant, what becomes scarce, what becomes newly worth doing because the cost of trying has fallen, and who helps organizations cross the gap between generic capability and real-world usefulness.
My bet is that the durable jobs of the AI era cluster around judgment, trust, customization, and coordination. Not because AI leaves those untouched. But because AI makes everything around them faster, cheaper, and more abundant — which is exactly what increases the value of the people who can direct, verify, personalize, and operationalize that abundance.
The first-order story of AI is automation. The second-order story is expansion.
And that is how big technologies create work: not by preserving the old task, but by making new layers of activity worth building. The real action is rarely at the first domino. It is in the pressure that moves after it falls.