What stays scarce when AI makes output cheap

Every month, more of what your business produces can be generated by anyone. The estimate, the follow-up sequence, the intake summary, the monthly report: all of it is heading toward the price of electricity. Which raises the owner question underneath the tool question. If output is cheap for you and equally cheap for every competitor, what do you still hold that stays scarce? Reading Emad Mostaque’s The Last Economy as an operator, I come away with a three-item inventory: the objective, meaning you can say what the work is for; attention, meaning you decide where human judgment actually lands; and position, meaning you own the routes to your customers, your records, and your recovery. None of the three gets cheaper when cognition does, and each tracks a specific chapter of the book, so I will take them one at a time.

The objective is yours to write

Mostaque’s Chapter 15, “The Alignment Economy,” makes the argument that moves first: once intelligent action becomes cheap, the constraint moves up a layer. The scarce capacity stops being the ability to do more work and becomes the ability to specify what the work is for. You can feel this in any automation you have already bought. An agent can optimize appointment fill, response time, or collections speed harder than any human manager ever did. What it cannot do is decide whether that target deserves optimizing.

And a shallow target produces a shallow win. A clinic improves collections by letting billing get more aggressive with the wrong patients. A service company improves utilization by quietly overloading its best technician. A sales team raises close rate by approving weak-fit customers who create chaos for the next six months. In each case the automation performed; the business got worse, because nobody wrote down what the win was not allowed to cost. Our rule is that an automation path is not serious until the owner can state three things in plain language: the target with its horizon, the values the target may not trade away, and the wins that would count as failures. Writing that contract is the job that cannot be delegated to the vendor, because the vendor does not carry the loss when the objective is wrong.

Attention does not scale

Chapter 21 names the second asset: attention is the scarce resource that remains after computation gets cheap. Once software can produce endless drafts, options, reminders, and recommendations, the binding input in your company becomes who reviews the exception, who decides which edge case matters, who changes the rule, and who turns one good answer into durable company memory. Those are attention acts; they do not scale, and most AI rollouts tax them instead of protecting them. The failure has a shape worth naming: an automation adds a review surface without reducing any ambiguity, so you now read more low-value summaries and clear more soft escalations than before. That is not leverage. I think of it as attention debt, a faster claim on the same person.

So run a keep-or-kill test on every workflow you have added. What human judgment does it protect? What durable record or rule improves when that judgment gets applied? What attention does it no longer need from you or your senior operator? A workflow with strong answers is compounding. A workflow with weak answers to all three is producing churn, however impressive its volume, and killing it returns the scarcest thing you have.

Position beats effort when the route is rented

The third asset is structural, and Mostaque gives it the darkest chapter title, “The Network Prison.” Chapter 10’s argument: leverage follows position in the network, and when the routes to customers, data, or coordination run through one hub, effort improves the hub’s position before the participant’s. The small-business version is easy to live inside without noticing. New work arrives through a marketplace profile. Customer history lives in a vendor-owned app. Follow-up runs through one messaging channel. The operation looks modern and efficient right up until the ranking drops, the pricing changes, or the account gets constrained, and then a strong operator with excellent tools discovers that the business was reachable only by permission.

Three routes decide whether that can happen to you. The route to customers: first-party lists, consented follow-up paths, a web property you own, and a record of why customers come back. The route to operating memory: job history, pricing context, and proof of work that stay legible outside the vendor’s app, because an export nobody can reconstruct a workflow from is storage, not memory. The route to recovery: a manual fallback, an alternate provider, a review queue with enough context to continue when an integration breaks. None of this means leaving the platforms; most small businesses should keep using the marketplace, the vertical SaaS, and the model API. The line to hold is between convenience and control, and it gets priced on one day: the day the platform turns. Convenience is worth nothing that morning. A reachable customer list, a legible record, and a rerouteable workflow are worth the business.

Your reflexes were trained on the old scarcity

The harder part is that your own management habits were built for the world where expertise was the expensive input. Mostaque’s Chapter 3, “The Seven Fatal Lies of a Dying Paradigm,” argues that scarcity, labor value, equilibrium, money-as-value, and distribution-by-contribution stop being safe assumptions once intelligence becomes abundant. Read as a mirror rather than a macro lecture, four of those show up in a small company as daily reflexes. Rationing access to the smart person, when the better move is making the standard and the exception rule visible enough that more people and more agents can work inside them safely. Pricing people by visible busyness, when the valuable operator is now the one who defines the method, supervises the agents, and catches the false positive. Waiting for workflows to settle on their own, when cheap tools amplify whatever design they inherit, so a weak quoting rule just becomes faster weak quoting. And reading rising output volume as rising value, which is the reflex I trust least; I wrote separately about which numbers still mean something in an AI-run business, because volume is now the easiest number to inflate.

The scoreboards around you still price labor

There is an outside version of the same problem. Mostaque argues that abundance can read as failure inside institutions built for scarcity, because those institutions score value mainly through human labor demand: fewer hours look like lost income, fewer intermediaries look like shrinking opportunity. A lender, a development program, or a jobs metric can look at a business that needs fewer coordination hours per unit of competence and score it as contraction, even when the remaining human roles moved up into design, oversight, and ownership. Expect the misreading; do not build for it. The tempting response is make-work preservation, review steps nobody needs and approval loops that survive only to justify roles, and it keeps people busy while stripping their leverage. The owner questions run the other way. Did the workflow remove drudgery, or did it remove the path by which a capable person learns the business? Did it reduce founder dependence, or centralize control in a vendor? Is the company more governable, or the same company with fewer people allowed to matter?

One honest limit before the summary. The Last Economy argues at macro scale, and whether my reading of it as a small-business asset inventory holds is an open question; I have not tested it against operating data, and a business whose real bottleneck is licensed labor or capital equipment may find its scarcity somewhere else entirely. So check the inventory against your own constraint. If a platform outage would hurt you most this quarter, work on routes. If your automations keep escalating to you, work on attention. If your agents are optimizing hard toward a target you never quite chose, the objective comes first. The one strategy I would rule out is competing on output volume, because that is the single game where the tools guarantee you no edge. Write the objective yourself, spend attention only where judgment changes durable state, and hold the routes. Cheap output took away one kind of advantage. What it left behind is everything about the business that was actually yours.

---

Source evidence used in this note: Emad Mostaque, The Last Economy, for the arguments that specification becomes the constraint once intelligent action is cheap (Chapter 15, “The Alignment Economy”), that attention is the scarce resource remaining after cheap computation (Chapter 21), that network position decides who keeps leverage (Chapter 10, “The Network Prison”), that scarcity-era assumptions stop being safe defaults (Chapter 3, “The Seven Fatal Lies of a Dying Paradigm”), and that abundance can read as failure inside institutions that score value through labor demand. Hadto interpretation: the three-asset inventory, the objective-authoring contract, the attention test, the route audit, and the capability-over-output questions are operating judgments translated from the book’s macro claims, and they have not been tested against operating data.