Scarcity migration

Cognitive and creative production is ceasing to be the constraint, and the economy is beginning to reorganise around what remains scarce: taste, judgment, trust, and the willingness to commit to a specific group of people and their specific problems, rather than building for everyone and therefore for no one.
For twenty-odd years, the venture capital gospel was that ideas are cheap and execution is all that matters. As the saying went, a thousand people have your idea in the shower every morning, and the one who wins is the one who can ship it. Paul Graham preached it, and Marc Andreessen built a firm around it. The entire scaffolding of the startup economy rested on the assumption that building things is hard and thinking of things to build is easy. That paradigm is now inverted.
A solo operator with the right tools can now ship a working product in a weekend that would have taken a funded team six months in 2019. Nor are we talking about a toy or prototype. This is a proper thing with authentication, payments, a design system, and deployment.
So the bottleneck has moved. Execution—the part that used to devour the budget, calendar, and occasionally the founder’s marriage—is being compressed into a commodity. What remains scarce is the original perception, the recognition that a specific group of people have a real unmet need, and the taste to solve it in a way that earns trust. The hard part is no longer building. The hard part is knowing what to build, and for whom, and why they should believe in you.
This feels like and perhaps is a seismic shift, but it is not unprecedented. Scarcity has migrated before. In fact, every major economic era can be read as a story about what was scarce, where that scarcity moved, and how long companies and institutions took to respond.
What Galbraith saw
John Kenneth Galbraith published The Affluent Society in 1958 and made an observation that offended almost everyone. In his view, America had solved the production problem. Factories could produce more goods than people naturally wanted, and the economy responded not by easing off but by creating desire.
Advertising existed to create the wants that production then satisfied, in a loop he called the “dependence effect”. The scarcity of goods was over. Scarcity of demand took its place, and an entire industry grew to paper over the gap.
Galbraith followed in 1967 with The New Industrial State, where he described the technostructure: a layer of managers, engineers, planners, and specialists that large firms needed to coordinate the complexity of modern production. The corporation was big because execution was complex. The technostructure existed because no single person could hold the whole machine in their head. Size was not greed or bloat (although it could be both); it was generally a rational organisational response to the difficulty of getting things done at scale.
What AI is doing to cognitive and creative production now mirrors what postwar industry did to physical goods. Traditional constraints are dissolving. And just as Galbraith’s America had to reckon with an economy organised around a scarcity that no longer existed, we must reckon with an economy organised around the difficulty of execution at a moment when execution is becoming trivially cheap.
The economics of the flip
Ronald Coase asked in 1937 why firms exist at all, given that markets are supposed to allocate resources efficiently. His answer was transaction costs. It is expensive to find suppliers, negotiate contracts, monitor quality, and enforce agreements, so firms bring those functions inside their walls, where hierarchy can manage them more cheaply. The boundary of the firm sits where the cost of doing one more thing internally equals the cost of buying it on the market.
AI is driving a sledgehammer through those transaction costs. Code generation, design, copywriting, data analysis, legal review, and customer support can now each be outsourced to a model at a fraction of the cost and delay of hiring someone or briefing an agency. The Coasean boundary of the firm is shrinking faster than handwriting on a birthday card when you started too big and now have to fit “and all the best for the year ahead” into a space the size of a stamp.
What once required a department can now be handled by a subscription. What once required a company can now be handled by one person with a laptop and a few API keys. The organisational unit of production is shifting toward the individual, not because individuals have become smarter, but because the cost of coordinating with machines has fallen below the cost of coordinating with people.
William Baumol identified in the 1960s that certain sectors—education, healthcare, the performing arts—resist productivity gains because their output depends on human labour that cannot be sped up. A string quartet cannot play faster. Well, yes, if we’re being pedantic, they can technically play a higher tempo, but you still need the same number of musicians.
Baumol’s cost disease meant these sectors became relatively more expensive over time as productivity gains elsewhere pushed wages up. AI is now targeting these sectors—the ones that resisted automation for sixty years: legal drafting, medical triage, tutoring, and content production. The string quartet still needs four musicians, but the administrative apparatus around it—booking, marketing, programme notes, grant applications—can now be handled by a model that charges by the token.
Paul Romer argued in his 1990 growth theory that ideas are non-rival goods. In other words, my use of an idea does not diminish yours. This quality makes ideas the engine of endogenous growth, and Romer won a Nobel for formalising this. But non-rival also means non-excludable in practice, and that is the crux. When execution was the moat, a good idea was protected by how hard it was to build. You could see what a company was doing and still not be able to replicate it, because doing so required a hundred engineers, a regulatory team, and an advertising war chest.
When execution compresses to near zero, an idea becomes copyable as soon as it becomes visible, and defensibility shifts. It moves from the ability to build to the ability to occupy a market, accumulate trust, own distribution, and compound a relationship that a late entrant cannot shortcut. The moat was once technical, but increasingly it is relational.
The niche dividend
Chris Anderson’s 2004 long tail thesis argued that the internet would unlock markets for products too niche for physical shelf space. He was right about distribution. A printed book that would never survive in a Waterstones could still sell steadily through Amazon. But production cost remained a constraint. For anything of quality, you needed a decent author, an editor, and perhaps a designer and a typesetter. The long tail of distribution met the stubborn floor of creation cost, and the tail stayed thinner than the theory promised.
That floor is now finally collapsing. When a solo operator can build, ship, and maintain a software product for a few hundred users at near-zero marginal cost, markets that were never economically viable start to open. A scheduling tool for independent dog groomers. A compliance tracker for small-batch distilleries. A booking system for narrowboat holiday lets. None of these justify a venture funding round. None of them justify a ten-person team. But they might justify one person spending a weekend, and if each niche sustains a livelihood, the aggregate reshapes the economy from below.
The interesting consequence is distributional. The venture model favours large addressable markets because venture economics depend on massive returns from a few bets. When production costs fall to near zero, the minimum viable market falls with them, and the range of problems worth solving expands dramatically. An economy of a thousand niches looks nothing like our current winner-take-all platform economy. It is more artisanal and more fragmented, which, depending on your politics, is either a return to the workshop or a race to subsistence. Probably both.
The last scarcity to fall
With production costs collapsing, the niche dividend means markets of 500 people are now viable. But a viable product still has to find its people, and for most of the internet’s history that has been the remaining bottleneck—the part you still needed money to solve. That constraint is thinning, too.
Galbraith’s dependence effect—manufacturing demand to absorb production—gets a turbocharged sequel. Advertising has become more targeted for decades, and the trend points toward almost perfect transactional reach. You can find your 500 potential customers with a budget measured in thousands of pounds rather than hundreds of thousands, because the ad platforms’ own AI will know who those people are before you do.
Herbert Simon observed in 1971 that a wealth of information creates a poverty of attention. That observation has aged well. But when reach becomes cheap, and attention remains expensive, the question flips from “how do I find my audience?” to “why should my audience pay attention to me?” Distribution, the second-to-last scarcity, is dissolving. What remains is trust, taste and reputation.
Follow the chain all the way down. Production is cheap. Distribution to a niche is cheap. Reach is becoming increasingly transactional. Every historical constraint in the sequence from idea to customer is compressing into a commodity. So what, if anything, is still hard?
Where scarcity remains
Three things remain scarce, and they are the things that no model has yet compressed.
The first is sound judgment, and its less obvious twin is focus. Ideas were never the scarce input; validation was. The shower is full of ideas. Most of them are bad, and the ones that aren’t could be bad for you specifically unless you have the domain knowledge, relationships, or positioning to execute them credibly. Cheap execution gives you cheap testing. You can build and ship a version in a weekend, discover that nobody wants it, and walk away having lost a weekend rather than a year. The scarce resource isn’t the idea, but the capacity to choose the right one and kill the wrong ones quickly.
But selection is only half the discipline. The other half is refusal. When you can build anything easily and cheaply, the temptation is to build everything, and those who give in end up with products bloated with features that seemed like a good idea at the time or met some idiosyncratic customer need. The ability to look at a viable idea and say “not this, not now, not us” is much harder when the build cost is zero, because the cost no longer enforces the constraint for you. Focus used to be imposed by scarcity. Now it has to be chosen, which is a fundamentally different demand on the person making the calls.
Michael Polanyi’s work on tacit knowledge sharpens this further. You often can’t articulate what you know until you try to build it, and the building reveals the knowing. Cheap execution speeds up that feedback loop, but it does not replace the knowledge behind it. The valuable judgment is not the kind you can write down. It is the kind you build by failing at the right speed.
The second is trust. George Akerlof’s market for lemons applies with force when production cost stops signalling quality. If anyone can ship a basic SaaS product in a weekend, many people will, and most of what they ship will be mediocre, unmaintained, and abandoned within six months. The buyer’s problem shifts from “can I find a solution?” to “can I find a solution that will still exist next year?” Provenance, track record, visible commitment to a market — the things that take time to accumulate and cannot be faked by a model — become a moat. The flood of cheap production paradoxically re-financialises quality. Being ok is table stakes. Being trustworthy is the differentiator.
This connects directly to the distribution question above. When reach is transactional, but attention is increasingly scarce, the only signal that breaks through is reputation—earned by showing up consistently in a specific market and being right more often than not. Cheap reach without earned trust is a loudspeaker in an empty field.
The third is organisational, and it comes from the consulting angle rather than the MBA reading list. Software execution is increasingly solved; execution inside organisations is not. Integration, compliance, change management, the politics of a twelve-person leadership team that cannot agree on what the software should do, and a regulatory environment that changes requirements halfway through the build all get in the way. The last mile of implementation remains stubbornly human, because the last mile is not a technical problem; it is a social one. Anyone selling the idea that AI dissolves all execution friction is either lying or has never tried to get a procurement team at a local authority to sign off a new vendor.
The uncomfortable implication is that most traditional companies are not well-run machines temporarily held back by expensive execution. They are inefficient, often bordering on dysfunctional organisations, where misaligned incentives, siloed teams, and institutional inertia absorb most of the energy that is nominally directed at getting things done. Cheap AI tooling does not fix this. It makes it more visible, because when the technical excuses disappear, organisational decrepitude has nowhere left to hide.
What those companies will look like in five years deserves its own article, but the short version is that Coasean logic cuts both ways. If the boundary of the firm shrinks because transaction costs fall, then large companies survive only if they can truly do something a network of smaller operators cannot — things like coordination at scale, regulatory moats, or capital-intensive infrastructure. For the many firms whose advantage was never much more than “we already have the team assembled”, that advantage fades quickly.
What follows
If that all sounds depressing, it need not be. Each remaining scarcity points to something practical.
If judgment is the scarce input, then the skill is not prompting or building, but selecting. People who succeed will treat every project as a probe rather than a product—ship, watch what happens, and build the pattern recognition that helps them spot a viable niche before the data confirms it. This discipline is not creativity. It is the willingness to throw away what you just built because the signal said no.
If trust is the moat, then owned distribution compounds and rented reach does not. An email list, a reputation in a specific community, and a body of published thinking that people return to because it has been right before are assets that survive platform churn. The advantage is not technical. It is relational, built on years of presence in a market and a clear, credible, visible commitment to it.
If the organisational last mile resists automation, then the consultancy that helps a company figure out what it needs before handing the build to a model has a durable role, even (or especially) in a world where the build itself is trivial. Knowing what to build, within the messy constraints of an organisation with competing priorities and legacy systems, is a different kind of judgment from niche-spotting. It is slower, more political, and just as resistant to automation.
The Galbraith turn
Galbraith died in 2006, just before the inversion his framework predicted. He understood that production had ceased to be the constraint and that the economy had reorganised itself around creating and manufacturing demand. He would have recognised at once what is happening now, the next turn of the same wheel. Cognitive and creative production is ceasing to be the constraint, and the economy is beginning to reorganise around what remains scarce: taste, judgment, trust, and the willingness to commit to a specific group of people and their specific problems, rather than building for everyone and therefore for no one.
The hard work has not disappeared. It has migrated, as it always does, to where the machines have not yet followed. The people who notice where it has gone, and get there first, will build the things that last. Everyone else will build a thousand apps in a thousand weekends and wonder why nobody cared.
Wondering if we’re a fit for your project? Ask your AI. It’ll read up on us, ask you a few questions and start the conversation.
More from the blog
-

Chaos or death
AI has made disorder cheap to create. Order still costs leadership attention and decisions, and most firms buy only the tools.I spend a lot of time in other people’s messes. That is, more or less, my job. A company calls because its technology initiative has stalled, and I always find the same scene, just in a different company. Everyone agreed on a strategy deck, and nobody followed it. Staff are using AI tools the business never chose, and Harmonic Security tracked 665 of them across the enterprises it studied in 2025. The person who commissioned it is frustrated that people aren’t follow…
-

Workslop: worse but more
In November 1955, a naval historian named C. Northcote Parkinson opened a satirical essay in The Economist with a line that outlasted everything else he wrote. Work expands to fill the time available for its completion. His example was an elderly lady who spent an entire day posting a single card to her niece, a task a busy person would finish in three minutes on the way to lunch.Parkinson’s law now has a sequel, and it is worse. Work no longer expands to fill the time. It expands to fill the capacity. The effort of producing a thing used to carry an implicit question, namely, whether the thi…
-

Foomers vs doomers: skilled incompetence and the AI tug of war
foom-er, noun, an ironic malapropism playing on doomerism (AGI fatalists) and FOMO (hustle bros), used to describe techno-optimists. Reflects the onomatopoeic “slowly, then all at once” nature of exponential growth: “AI go foom.”A speaker at one of the many AI conferences that have sprung up like garments on your “clothes chair” (not clean, but not dirty enough to wash) describes the miracles of agent-assisted coding. Year-long rewrites finished in a fortnight, backlogs cleared, and hard problems dissolved by brilliant prompting and clever code harnesses. The audience of true believers laps i…
-

Forward Deployed Engineers: AI’s hired guns
The last mile of AI deployment is a real person, sitting in your office, learning your business, writing your plumbing, and making things work reliably. That will remain human for the foreseeable future.Google Trends shows a 5,000% increase in searches for “AI forward-deployed engineer” over the past five years. The curve stays flat from 2021 through mid-2024, then turns vertical. Indeed data reported by Business Insider shows postings for the role rising from 643 in April 2025 to 5,330 a year later, a 729% increase. LinkedIn’s own study found 42-fold growth between 2023 and 2025, making fo…
-

The sorting machine: recruitment’s race to the bottom
The sorting machine was built to cope with a genuine problem. Where it went wrong was in mistaking efficiency for effectiveness, in assuming that the ability to process resumes fast meant the ability to evaluate candidates well. The fix is not a smarter machine but a more modest one, paired with humans who are given the time, the tools, and the structured information to do what they were always better at.If you’ve been involved in any part of the recruitment process, the last 18 months have been dispiriting, to say the least. Hiring became a closed loop. Employers adopted applicant tracking…