Cheap Thinking Makes Strategy Harder

Here is how the plan used to get made, and it deserves to be described properly rather than caricatured.

You watched the industry. You went to the events. You saw which competitors were hiring and which had gone quiet, and you knew roughly what was coming in the next release cycle because you had watched the last eight. Then you did the arithmetic. You make your team ten per cent bigger, you aim for eleven or twelve per cent more revenue, and away you go. Year on year, more profitable, stronger.

That is not naive. It is a method, and it was calibrated — tuned against a world that changed slowly enough for tuning to mean something. The people who ran it for twenty years were not fools, and a book that begins by implying otherwise has misdiagnosed the problem in its first paragraph.

So what actually broke?

The method never claimed to predict the future. It claimed something weaker and far more useful: that the set of things which could plausibly happen next year was small enough to hold in a room. Competitors were enumerable — you could name them. Product cycles were observable. Customers changed at the speed of procurement. The plan was extrapolation plus a short list of contingencies, and the list stayed short because reality kept it short.

That precondition is what expired. Not the rigour, not the people, not the quality of the analysis. The plans got better and the ground they stood on got smaller.

Most explanations for the current disorientation point somewhere else. They point at speed. Everything is faster, the tools change every quarter, the release cadence is relentless, keep up. That explanation is popular for a reason: speed is answerable by running harder, and running harder is something a firm knows how to organise. Set a target, appoint an owner, buy a tool, report monthly.

It is also the smaller half of the story, and mistaking the half for the whole is why so many well-run AI programmes are producing a strange, specific feeling in the people who commissioned them — that the organisation is measurably faster and the decisions are measurably less obvious.

AI makes thinking cheaper at exactly the same time that it makes strategy harder. Those are not contradictory. They are causally connected.

The load-bearing word in that sentence is causally. This is not the observation that AI creates uncertainty — every significant technology does that, and the observation is worth nothing on its own. It is the claim that the specific thing making strategy harder is the same thing your AI programme is celebrating internally. The cost of thinking fell. Everyone’s cost of thinking fell. And the consequence of universally cheap cognition is not convergence on the right answer; it is an explosion in the number of answers that are now affordable to try.

If your business is the floor

There is a blunter way to say the first half of that, and it is worth saying blunt before it gets precise. Thinking is now table stakes. It is the floor. And if your business is the floor, you’re out of business.

That line is directionally right and exactly one word short. The missing word matters enough that the next chapter is about supplying it, because the version without it produces a claim this book does not make — that judgment has been commoditised, that expertise is finished, that human effort is now decorative. None of that is true, and the argument is stronger without the overreach.

What this book is, and what it is not

It is not a forecast. No date appears in it — not for model capability, not for adoption, not for the end of anything. It is not a tool review. And it is not a re-teaching of work we have already published: the AI Fog belongs to The Terminal Value Doctrine , the economics of cheap cognition to The Great Reset , and disciplined search to The Cognition Dimension Ladder . Each gets referenced, none gets repeated.

What this book supplies is the join between them — the causal reason those three things belong in one argument rather than three adjacent ones. It answers a question that has been sitting unasked in a lot of boardrooms:

Why does access to vastly more intelligence leave my strategic choices less clear — and what kind of reasoning still works?

By the end you will be able to name what your invoice is denominated in and say whether cheap cognition compresses it; choose a boundary case that is worth a morning of your most expensive people; and close one entry in a search ledger with a decision that actually changed. That is a smaller promise than most books in this genre make. It is also one you can check.

The arithmetic arrives in Chapter 6. The instrument arrives in Chapter 10. First, the word missing from the floor.

Forty times per year. That is how fast the price of one fixed capability fell.

Epoch AI took six benchmarks and asked a question most cost commentary skips: not “how much does the best model cost?” but “how much does it cost to buy a fixed level of performance, over time?” The price of matching GPT-4 on a set of PhD-level science questions fell by roughly 40× per year. Across all the milestones they measured, the decline ran from 9× to 900× per year — and Epoch flag, in their own words, that the fastest drops occurred most recently, so it is “less clear that those will persist.”¹

Read the range rather than the headline. Nine to nine hundred is not a wave arriving evenly across a coastline; it is a series of specific tasks becoming nearly free at different moments, for reasons that have to do with benchmark structure and model architecture rather than with anyone’s business plan. That is the operationally important fact and it is almost always skipped.

It means a firm cannot schedule this. “We’ll deal with it when it arrives” is not a plan, because it has already arrived somewhere in your delivery stack and not somewhere else, and nobody sent a notification. The forward view is similar in character: Gartner project that inference on a trillion-parameter model will cost providers more than 90 per cent less in 2030 than in 2025² — a named forecast from a named house, which is a different kind of object from a measurement and is used here only as a direction of travel.

So what exactly became cheap?

Not thinking. Generic cognition.

The distinction is the whole chapter, and it is not a hedge — it is what makes the claim usable. “AI makes thinking cheap” is a slogan that can be agreed with and forgotten. “Reproducible cognition is becoming the floor, and your invoice is denominated in it” is a diagnosis somebody can act on this quarter.

Generic cognition is work that has a known shape before anyone starts it. The market scan. The three competitor profiles. The first-pass financial model. The standard memo. The compliance draft. The literature review. The status report that summarises the status reports.

Notice what the test is not. It is not difficulty — plenty of this work is genuinely hard and takes trained people years to do well. The test is reproducibility: could a competent stranger, given the same brief and the same inputs, produce a materially equivalent output? If yes, the work has a shape, and shaped work is exactly what the price collapse is eating.

That right-hand column is not our invention; it is where The Great Reset argued scarcity migrates once thinking stops being the constraint. It is quoted here for one purpose only: to bound the claim. This book does not argue that judgment has been commoditised, that expertise is finished, or that human effort is now decorative. The argument is narrower and harder to dismiss.

The anaesthetic version, and the uncomfortable one

“Judgment stays scarce” is true, and it is also the most widely used anaesthetic in professional services. It is repeated at conferences by people whose firms bill for exactly the category on the left.

Here is the uncomfortable version. The fact that your judgment is scarce does not save your invoice. If most of what you charge for is generic cognition, then the scarcity has moved and your pricing has not. Your judgment may be worth more than it has ever been worth, and your revenue can still fall, because the two are attached to different line items and only one of them is being compressed.

That gap between where the value sits and where the invoice sits is the subject of Part II. It is also the reason the argument cannot be answered with better thinking.

The constraint has already moved twice

There is a second-order version of this that matters for what comes later. The familiar move is from execution to decision: when producing things gets cheap, choosing what to produce becomes the bottleneck. That is correct, and it is where most of the industry currently sits.

The Cognition Dimension Ladder makes the further move: once choosing can itself be run against an evaluation function, the constraint shifts again — from the decision to the apparatus that produces decisions. Two shifts, not one.

Its only job in this chapter is to make a later argument available. If choosing is itself becoming automatable, then “we’ll compete on better thinking” is not yet a strategy. It is a location — and the location is contested. Chapter 5 needs that; hold it.

The first thing to go and do

What fraction of what you bill for has a known shape before you start?

This book will not invent that number for you, because it is firm-specific and any figure offered here would be decoration. But most firms can estimate it in an afternoon: take the last twenty engagements, decompose them into deliverables, and mark each one against the reproducibility test. Do it with the delivery leads rather than the partners, because the partners remember the interesting parts and the delivery leads remember the hours.

The number you get is your exposure. It is the first of three things this book asks you to do; the second is in Chapter 5 and the third in Chapter 13.

So: generic cognition is the floor, the floor is arriving unevenly, and the scarce thing has moved somewhere your pricing may not have followed. All of that is about your firm and your costs, which is how the story is normally told.

It is the smaller half. Everyone else’s cost of thinking fell at the same moment — and that is where the fog comes from.

We have a name for the condition, and the name has been doing too much work.

The AI Fog is the simultaneous compression of the credible forecast horizon and expansion of the plausible solution space. The visible distance shrinks; the map enlarges. Less time, more possibility, less clarity. That definition belongs to The Terminal Value Doctrine, which teaches it properly. This chapter does something narrower and, for our purposes, more useful: it pulls the two halves apart and asks where the second one comes from.

Term one: everybody is already managing it

Horizon compression is the half that has reached the boardroom. Half of CEO planning effort now sits inside horizons of less than a year, up from 43 per cent the year before³. Boards feel this directly — the three-year forecast that nobody in the room believes, the roadmap whose top quarter is fluent and whose bottom three quarters dissolve into adjectives.

And the standard response to it is correct. Shorter cycles. More frequent re-forecasting. Decide later, commit less. If horizon compression were the only thing happening, that response would be sufficient, and a firm that adopted it would be fine.

This book is not about that term. It is well covered, well understood, and already being managed by people who know how.

Term two: nobody manages it, because it looks like weather

Solution-space expansion gets treated as a background condition. Something in the air. The market is turbulent, disruption is everywhere, the pace of change is unprecedented — all sentences that describe a climate and imply no action beyond resilience.

That is the mistake this book exists to correct. The expansion is not weather. It has a source with a direction, and the source is the cost collapse Chapter 2 measured.

Which produces the claim that everything after this rests on: the Fog is partly endogenous. Cheap cognition is not only the instrument you use to see through it. It is one of the things making it. Your own cost collapse, and your competitors’, and your customers’, are inputs to the condition you are trying to navigate.

Cheap thinking does not make the future easier to understand. Cheap thinking creates more futures to understand.

Why splitting the term is operational, not pedantic

Take each half on its own and ask what it would demand. The answers are different, and they are incompatible in a way that matters.

Run them together and both prescriptions fail for the same structural reason: each one spends the resource the other needs. Speed spends evaluation. Optionality spends capital that has to be committed while the horizon is at its shortest. What survives is neither speed nor breadth. It is selection — deciding which question you are going to pay to answer.

That sentence travels. It works in strategy, and it works unchanged in security, in product design and in public policy — anywhere a fall in the cost of producing answers has outpaced any improvement in deciding which ones were worth producing.

The asymmetry that turns a condition into a paradox

Here is the part that should change how a leadership team reads its own AI progress report.

Your cheap cognition compounds inside your walls, at the speed of your change programme. It is gated by adoption, by training, by data access, by the security review, by the two teams that have not started yet, and by the general fact that organisations absorb new tools at the pace of their slowest necessary approval.

Everyone else’s compounds across the whole market at once. No adoption curve applies to the aggregate. No internal resistance slows it. It does not need your permission, your budget cycle or your change-management plan.

Which means a firm can have the best internal AI adoption in its sector and still be losing ground, because the outside term is growing faster than the inside one. The comparison that feels natural — us today versus us last year — is the wrong comparison, and it is the only one most dashboards can make.

There is a version of this argument that stops here, at the level of the felt condition, and it is the version that gets called insightful and produces nothing. If the expansion has a source, it travels routes — specific, nameable, and different enough from each other that a firm can be exposed to one and immune to another.

There are three that cheap cognition drives directly: customers, competitors, constructors. The next chapter draws each one separately, and starts in the room where the first one shows up.

The client opens a folder and puts something on the table before you have finished sitting down.

It is their own baseline analysis. Market sizing, three competitor profiles, a first-pass model with the assumptions listed down the side. It is not as good as yours. It is good enough that the conversation now starts at a point you used to bill for arriving at — and they are pleased with it, because it took a weekend and cost nothing.

Nobody in that room experiences this as a competitive event. No competitor was mentioned. No deal was lost. That is precisely why it matters, and it is why the customer channel is the one incumbents notice last.

Channel one — customers

The mechanism: your customer’s cost of thinking fell by the same order of magnitude as yours, and they spent it on the part of your work they most resented paying for.

The demand-side evidence is clearest where the unit of sale is most visible. Research cited by Clio finds that 67 per cent of corporate legal departments and 55 per cent of law firms expect AI to change how hours are billed, that 71 per cent of buyers already prefer a flat fee for an entire matter, and — the sentence that matters most — that “clients have moved faster than the industry has.”⁴

Read the direction of that gap out loud, because it inverts the usual disruption story: the demand side repriced before the supply side decided to. Not a startup. Not a platform. The people already paying you.

What expands here is not the number of competitors. It is the number of ways the customer can meet their own need — do it internally, do a rough version and buy only the hard part, buy the hard part from someone smaller, or decide the question was not worth the engagement after all. Each one is a future in which some portion of your revenue does not exist.

And there is a second-order effect that is easy to miss in the moment. A customer who has produced a first pass has also produced an opinion — about scope, about duration, about what this should cost. You are no longer proposing into a blank page. You are negotiating against a document you did not write.

Channel two — competitors

The mechanism: they did not get better at your business. They got cheaper at changing theirs.

Designing an offer, modelling its economics, testing it against a segment, specifying what would have to be built — all of that used to be expensive enough to ration. Firms attempted structural change once every few years because attempting it consumed the strategy function for a quarter. When the cost of attempting falls, the rate of attempts rises, and every attempt is another plausible future you now have to hold in mind.

Public markets have started pricing this. Sapphire Ventures’ software indices fell 20 per cent, and pure SaaS 23 per cent, through 18 February 2026 — with the sector decoupling from the broader market, IGV down 32 per cent while the Nasdaq was essentially flat⁵. They name the drivers directly: technology budgets shifting toward AI initiatives, pricing moving away from seats toward measurable value delivered, and continuous model improvement raising displacement risk for established incumbents. Their summary line is worth reading twice:

In short, risk is up, and terminal value assumptions are down (for now at least).

Include their hedge, because leaving it out would be the kind of citation this book criticises: Sapphire themselves say investors may be “selling first and asking questions later.” It may be an overshoot. The argument does not need it not to be. Even discounted heavily, a public market re-rating is the fastest-moving evidence available that the second Fog term has commercial consequences — it is, quite literally, a price being put on a set of futures.

Channel three — constructors

This is the channel where the argument is usually made badly, so it is worth making it against its own most convenient evidence.

The lazy version is “there are more startups now.” The data does not straightforwardly support that. Crunchbase recorded $300 billion invested across 6,000 startups globally in the first quarter of 2026, with AI taking $242 billion — 80 per cent of the total, against 55 per cent the year before. A record by a distance. But seed deal counts fell 30 per cent year over year to 3,800, and the dollar increase came “entirely due to larger rounds.”⁶

So the honest reading is that the constructor channel is not primarily about venture-funded startups. Fewer bets are being placed, at larger size, on fewer companies. If the argument rested on a rising count of funded entrants, the argument would be in trouble.

It rests on something else: the falling cost of construction available to everyone. The customer who decides to build rather than buy. The mid-sized competitor who can now specify and ship something that would previously have required a platform team. The two people who leave a firm with a clear view of one workflow and no legacy to protect. And you.

Which changes what the firm should model. The attacker to worry about is not necessarily funded, not necessarily a company, and not necessarily new. It may be your largest client’s internal team, who have just discovered they can do the first pass themselves — channel three arriving through channel one.

The chain, drawn once

That last point is the diagram’s whole content. Chapter 3 separated the Fog into two terms; the three channels feed only the second. A firm can shorten every planning cycle it has and not touch a single arrow in that figure.

So how many channels are there?

More than three. Regulators expand the space. Capital markets expand it. Platform shifts expand it, and so do supply chains and geopolitics. Those are real and they are somebody else’s chapter.

These three are here because they meet a specific test: you can trace the price of thought falling and see it cause the expansion, rather than merely coincide with it. A customer does the first pass because cognition got cheap. A competitor attempts a redesign because designing got cheap. A constructor builds because building got cheap. If you want to add a fourth channel, that is the bar — and a bounded claim with three defensible arrows is worth more than a comprehensive one with nine decorative ones.

Three channels, one destination, and an asymmetry that means internal excellence cannot close the gap on its own.

Which raises an uncomfortable question about the thing most firms are currently doing about it. If the outside term is growing faster than the inside one, what exactly is a productivity programme buying?

What a productivity programme buys, in most firms, is a dashboard that is entirely green.

Adoption at ninety per cent. Hours per deliverable down eighteen per cent. Turnaround times down. Rework down. Client satisfaction steady or slightly up. Not one red cell anywhere on the page, and the programme lead has earned every one of them — this is a well-run programme, delivered on budget, by people who did exactly what they were asked to do.

Hold that image, because the argument only works if the programme is genuinely good.

Having all your staff go ten per cent faster or twenty per cent more productive doesn’t get you through the Fog. That gets you to zero. That gets you out of business.

That is the conviction, and stated bare it is too strong. It needs a condition attached, and the condition is what makes it useful rather than merely alarming.

Read what the qualification adds. Productivity is not the enemy. There is no version of this argument in which working slowly is a strategy. The pairing is the problem — output gains bolted to a commercial unit that is losing its economic support, with nothing being built to replace the unit.

The three conditions

Whether your programme is buying margin or spending the firm comes down to three questions. They are checkable, they are specific to you, and none of them is about the technology.

Three out of three and productivity accelerates your depreciation. Two out of three and you have pressure without collapse — time to move, but a shortening runway. One or none and productivity is straightforwardly good: press the accelerator, hard, and enjoy it. This book is not written for that firm, but it is worth knowing you are that firm.

Same tool, opposite economics

The cleanest external statement of the mechanism comes from a vendor selling the tools, which is what makes it hard to argue with. Clio, on what AI does to a firm’s revenue: a tool that turns a four-hour task into a one-hour task “creates a revenue problem under hourly billing and a margin opportunity under fixed fees.”⁴

Same tool. Same efficiency gain. Same people, same clients, same quality. Opposite economics — decided entirely by what the invoice is denominated in, and not at all by the technology. Every argument about which model to buy is downstream of a question most firms have not asked.

Why nothing inside the firm can see it

The measurement instrument is the timesheet, and the timesheet counts the input that is shrinking.

Think about what that means for the reporting pack. A programme measured in hours saved is structurally incapable of reporting this harm, because the harm is denominated in the very thing being saved. Every hour removed appears twice as good news — once as efficiency, once as capacity released — and never once as revenue withdrawn. The instrument that would show it is revenue per engagement and the trend in the unit itself, which sits in the finance pack, in a different meeting, owned by someone who is not tracking the AI programme.

This is not a management failing. It is an instrumentation gap, and it is why the mechanism can run for years without anyone dishonest being involved.

Self-disintermediation, by accident

There is a name in our own doctrine for removing the intermediary layer your business model places between the customer and the value. The Terminal Value Doctrine calls it self-disintermediation, and treats it as something an incumbent should do deliberately — harvest the existing book, migrate the convertible assets, and construct the successor inside your own walls before an attacker builds it outside them.

What this chapter describes is the accidental version. The same structural move — the intermediary layer removed, the value handed to the customer — executed by a productivity programme, with none of the compensating construction, and no board decision at any point, because nobody proposed it. As Scott puts it: you’re disintermediating yourself.

The full version of that sentence opens the next chapter, because it deserves the arithmetic standing behind it.

The escape hatch, closed before you reach it

The natural response to all of this is to resolve to think harder. Get better at strategy. Point more cognition at the problem. It is the wrong move, and it is worth closing off now rather than in three chapters.

More ordinary thinking is not the answer. Undirected AI cognition produces still more options and makes the Fog worse — it is an option generator, and options are the thing there is already too much of. Recall the two-step constraint move from Chapter 2: the bottleneck is not the decision, it is the apparatus that produces decisions, and an apparatus without a rejection function is just a faster way of generating candidates.

Which is why Part III is about disciplined search rather than more search, and why it opens with a test for which questions are worth asking at all.

So stop the programme?

No — and this is the misreading that would do the most damage, so let it be explicit.

Stopping is worse. The compression happens anyway on the client’s side of the table, and a firm that has forgone the efficiency simply arrives at the same repricing conversation with a higher cost base. The instruction is to re-scope and re-label: keep the efficiency, move it out of the strategy column and into the operating budget where it belongs, tolerate it rather than celebrating it, and spend the freed strategic attention on the question the programme cannot answer — whether the unit survives.

All of which is assertion until someone puts numbers on it. So here is a firm, its unit, and a twenty per cent gain.

You use AI in your business, and now you’re charging 20% fewer hours for the same work. You didn’t reprice. You didn’t restructure. You did nothing except become more efficient. You’re disintermediating yourself. You’re cutting off your own arm. You’re charging the customer less, the customer’s getting all the benefit, and you’re helping yourself go out of business quicker. It is the exact wrong approach.

That is the claim. This chapter builds the instrument that tests it, and the next one runs the tests.

Before a single number, the honest framing — because a model that assumes away the obvious objections proves nothing:

Four escapes, named before the arithmetic starts. Chapter 7 tests each one rather than waving at it. This chapter’s job is narrower: build a firm precise enough that the tests mean something.

The firm, before

Forty consultants. Each delivers 1,000 billable hours a year, so the firm has 40,000 hours of annual capacity. The standard engagement takes 100 consultant-hours and bills at $300 an hour: $30,000 of revenue for a piece of work the firm has done hundreds of times.

At 400 engagements a year the firm consumes its full capacity and turns over $12.0 million. That is a practice at its practical ceiling, which is where a healthy professional firm lives — and note that being at the ceiling is normally read as a success signal.

On the cost side: fully loaded cost of $180,000 per consultant gives $7.2 million; overheads — premises, support staff, systems, insurance, partners’ drawings below the line — add $2.4 million. Total cost $9.6 million. Operating margin $2.4 million, or 20 per cent.

This is a good firm. Not a distressed one, not a badly run one, not one that has been ignoring technology. The mechanism about to be described does not need a weak firm, and that is the part worth sitting with: it bites hardest when the firm is well run, because a well-run firm is the one operating at capacity with its efficiency gains fully realised.

The slack that is already in the model

Before AI touches anything, notice how much leakage a labour-priced model carries as standard. In legal services — where the numbers are published rather than inferred — utilisation runs at around 38 per cent, realisation at 88 per cent and collection at 93 per cent, so that “law firms ultimately collect revenue for only a fraction of the workday.”⁷

Two things follow, and both matter for how the model should be read.

First, our composite assumes far better utilisation than that. It is generous to itself. If the mechanism hurts this firm, it hurts a typical one harder.

Second, the leak is multiplicative. Every hour worked passes through three filters — is it billable, is it billed, is it collected — before it becomes a dollar. AI compression attacks the numerator of all three at once, because it removes hours from the top of the funnel rather than improving conversion through it. And not one of those three filters appears on the AI programme’s dashboard.

The compression

The firm deploys AI properly — research, drafting, first-pass analysis, document review, the parts of the work with a known shape before anyone starts. The standard engagement falls from 100 hours to 80.

Twenty per cent is deliberately modest. Firms are reporting more, in the parts of the work where the compression has landed. The argument does not need more, and a conservative parameter is harder to dismiss.

Now the starting position, in the source’s own terms. Suppose a job used to require 100 consultant-hours and now requires 80. If the firm simply passes through 80 hours at the same hourly rate, it has used its own AI investment to shrink its revenue unit by 20 per cent. The customer has captured much of the cognition-cost collapse. The firm’s product hasn’t changed. Its pricing hasn’t changed. Its value proposition hasn’t changed.

The base case, run

Everything else held constant — same clients, same win rate, same people, same rate card.

Read it row by row, because the rows are the argument.

Hours delivered fell by 8,000. Those hours are not a saving. The firm still employs the people; the salaries are annual and human and were agreed before the model changed. In a labour-priced firm, freed capacity is not a cost reduction — it is unsold inventory, and it stays unsold unless something else fills it.

Utilisation falls to 80 per cent, and note that this is the optimistic reading, in which the firm keeps every one of its 400 clients and loses no work at all.

Cost does not move. That single row is the whole mechanism: the compression lands on revenue, and the cost base does not follow it down. There is no lever in the middle of the year that converts a 20 per cent efficiency gain into a 20 per cent cost reduction without doing something a partnership would recognise as a crisis.

Margin: 20 per cent to zero. A successful, on-budget AI programme moved the firm from a healthy margin to break-even — and every internal metric improved while it happened.

So where did the money go?

To the client, in full, without a negotiation. Nobody asked for a discount. The firm gave one by operating its billing model as designed.

And nothing in the firm's reporting will describe it that way. The AI programme reports hours saved, cycle time, adoption and satisfaction — all improving. The number that would show the harm is revenue per engagement, and the trend in the unit rather than the input, and that number does not live in the programme’s pack. It usually does not live in any single person’s pack: the efficiency sits with operations, the pricing sits with the commercial lead, and the trend across both sits with nobody.

The dashboard is green because the dashboard measures the thing that is shrinking.

Of course, no firm sits still through this. There were four escapes on the table before the arithmetic started, and each of them is a real option with a real condition attached.

So what happens when the firm reacts?

Escape one — fill the freed capacity

The firm has 40,000 hours and each engagement now takes 80 of them, so it can deliver 500 engagements instead of 400. At $24,000 each that is $12.0 million — revenue restored exactly, margin back to $2.4 million. On paper the problem disappears.

The condition is doing all the work, and it should be said plainly: demand must exist for 25 per cent more engagements, at a price that just fell. Not more demand in general. More demand for this firm’s work, this year, won in competition.

Then the sensitivity, which is what turns this from a hope into a test. Suppose the firm sells 440 rather than 500 — a 12 per cent shortfall against what it needs, which most partners would describe as a decent year.

A 12 per cent miss on volume costs 60 per cent of the profit. That is operating leverage doing what operating leverage does when the cost base is fixed and human: it runs both ways, and it is unforgiving on the way down.

There is also a competitive complication that no amount of sales effort touches. Every competitor in the market got the same 20 per cent compression in the same period, so every competitor’s effective capacity rose by the same 25 per cent. The firm is chasing 25 per cent more volume into a market whose supply just rose 25 per cent. That is a supply shock the sales team is being asked to absorb, and treating it as an activity problem burns a year finding out.

Escape two — hold the price per engagement

Bill $30,000 for 80 hours of work. Revenue holds at $12.0 million, margin returns to $2.4 million, and the effective rate rises from $300 to $375 an hour.

This escape works. Say it without hedging — the book is not arguing that professional firms are doomed, and a reader who finishes Part II believing that has been misled.

But look at what has actually happened. The firm has stopped selling hours. It is now selling a defined piece of work at a fixed price, which is a different commercial unit carrying different risk (scope, not time), different discipline (estimation becomes load-bearing), different competitive exposure (fixed prices are comparable in a way hourly rates are not) and a different failure mode (the engagement that runs long is now yours).

It is a good move. It is not a pricing tweak — and most firms that reach for it reach for it under pressure, without the delivery architecture a fixed price actually needs. Hold that thought; it is where the whole argument lands at the end of the next chapter.

Escape three — raise the rate

Arithmetically identical to escape two: $375 an hour, 80 hours, $30,000. Behaviourally, a different proposition entirely, and the difference is the client’s line of sight.

Under escape two the client buys an outcome and never prices the hours. Under escape three the client is asked to pay more per unit for a job that visibly took less time — and the invoice tells them so.

The market-level evidence suggests this is being competed away. The Thomson Reuters Institute’s Law Firm Rates Report 2026 finds that “regardless of whether firms discount aggressively or hold firm on realization, they’re collecting roughly the same amount per hour.”⁸ Whatever an individual firm’s posture, the yield per hour is converging — which is what a competitive market does to a price that is not attached to a differentiated unit.

Escape four — keep the dividend quietly

Deliver in 80 hours. Bill 100. Say nothing.

This one deserves to be taken seriously rather than dismissed, because it worked for thirty years and the people who ran it were not fools either. When Westlaw and LexisNexis compressed legal research from hours in a library to minutes at a terminal, “the efficiency dividends didn’t flow to clients. Firms captured it, and billing rates continued climbing.”⁹

If you had made this book’s argument in 1985, you would have been wrong for three decades. So what is different?

Two things, in the source’s own terms. The earlier tools “optimized around the lawyer… They made lawyers more productive but left the core work untouched.” Today’s tools “are beginning to do the work itself” — drafting, reviewing, analysing, summarising. And the client holds the same tools, so they no longer merely suspect the work got faster. They can produce a version and see.

Which gives the escape a shelf life rather than a verdict. In the short term AI may expand margins if pricing holds, and early adopters may benefit significantly — but “that window won’t stay open for long as prices adjust across the market over time.”

So escape four is real and temporary, which makes it a harvest rather than a strategy. The distinction that matters is not whether you take it — take it — but whether the harvest is funding something. Cash from the old book is exactly how the successor gets built. A firm that takes escape four and does nothing else has bought time and spent it.

What happens when the channels arrive together

Each escape was tested against one variable at a time. That is not how it happens. Chapter 4’s customer channel does not wait politely while the firm works through its pricing options.

Assume 30 per cent of the original 100-hour engagement was first-pass analysis — the scoping, the market scan, the baseline model, the competitor summary. (An illustrative share, and in the same range our own worked ledger entry uses for first-pass scope in an advisory firm.)

The client now does that themselves, because their cost of thinking fell too. The addressable work drops to 70 pre-compression hours; with the firm’s own AI applied to what remains, roughly 56 billable hours. At $300 an hour, the engagement is worth $16,800 — down 44 per cent from $30,000.

The last two rows are the finding.

To stand still, the firm must win 79 per cent more engagements — and at 714 engagements it is sitting exactly at capacity again, with no slack left for the next increment of compression. The escape hatch closes behind it.

If it wins no new work at all, revenue is $6.7 million against a $9.6 million cost base: a $2.9 million loss, from a firm that did nothing wrong operationally and hit every target in its AI programme.

The three hinges

Strip the four escapes down and three hinges remain. Each one decides whether the arithmetic above is benign or fatal, and the important column is the last.

Assume hinge two and you are forecasting, not reasoning — so make it an assumption with a name and go and test it against your own pipeline rather than against the industry press.

Which collapses the list. Escape two and hinge three are the same move: stop denominating the invoice in the thing that is being compressed. It is the only lever the firm owns outright, and it is not a pricing decision. It is a design question.

Before designing anything, though, the model deserves to be attacked properly — including by the strongest argument against it, which has thirty years of evidence behind it.

The last chapter answered the 1985 problem inside escape four. Now the whole model goes on trial, and the prosecution gets to open.

That objection deserves a real answer, and the answer starts with three concessions.

What has to be conceded

It is a model. Forty consultants, $300 an hour, 400 engagements — every one of those figures was chosen for legibility. No firm was measured. Anyone treating the $2.9 million loss in Chapter 7 as an empirical result has misread the chapter, and this book would rather lose the rhetorical force than have it quoted as data.

The elasticity is assumed. Hinge two is an input. A reader who believes their market will absorb 25 per cent more volume at a lower price can change it, and the model will agree with them — which is the point of a model with visible parameters rather than a conclusion with a chart.

It says nothing about when. There is no date in it and none will be supplied. A firm that wants a timeline is holding the wrong instrument, and the entire method in Part III is built on the assumption that timelines are unavailable.

What survives all three concessions

Something narrower, and sturdier: the sign of the derivative.

Under the three conditions named in Chapter 5, faster delivery of a labour-priced unit reduces revenue rather than increasing it. Not “may reduce” — reduces, by construction, unless one of the three hinges holds. The magnitude is uncertain. The timing is unknown. The direction is not in question, because it follows from the arithmetic of a unit that counts the thing being removed.

Two supporting claims survive with it.

The conditions are checkable. A firm can test all three against its own last twenty engagements in an afternoon — what does the invoice count; could the client produce a rough version; what fences the layer above. That is a rare property in strategic argument, which usually asks you to accept a worldview rather than run a test.

The firm’s instruments cannot see it. Chapter 6 made this point about the timesheet, and it generalises: any programme measured in units of the thing being removed will report the harm as a success. This is structural rather than a failure of diligence, and it is why the mechanism can run for years with nobody behaving badly.

Run the diagnostic on yourself

Three questions, from Chapter 5, restated as a scorecard rather than re-argued:

• Is the unit compressible? What does your invoice count — and is cheap cognition removing it?
• Is the compression visible or performable? Could your client produce a rough version of your deliverable this week?
• Is entry above open? What stops a credible outsider from selling the layer above you next year?

Three out of three and Part II is about you; the arithmetic applies and the only question is how long you have. Two out of three and you have pressure without collapse — a shortening runway rather than a cliff. One or none and you should go and press the accelerator, hard, and enjoy the decade. Knowing which of the three you are is worth more than agreeing with the book.

What would prove this wrong

A claim without falsifiers is not a claim, so here are four. Each is observable, and each would sink the argument if it held at market level rather than in one firm’s anecdote.

The limit on scope

Part II describes one firm shape: labour-priced professional services with a compressible unit. It is not a theory of all businesses, and reading it as one would be the book’s own version of the overreach it criticised in Chapter 2.

Firms whose advantage rests on regulatory position, physical assets, exclusive distribution or trust that took a decade to earn may be exposed through a different channel, at a different speed, or barely at all. Chapter 11 draws that boundary properly, with a test, and shows a domain where cheap cognition reduces uncertainty rather than multiplying it.

What generalises even where the conclusion does not is the question. Any organisation can ask what its invoice counts and whether cheap cognition is removing it. Most have never asked, because the invoice is treated as an accounting artefact rather than a strategic one.

Self-disintermediation, paid for out of the AI budget

Our own doctrine asks incumbents to remove their intermediary layer deliberately: harvest the existing book, migrate the convertible assets, construct the successor inside your own walls before someone builds it outside them.

The firm in Chapters 6 and 7 performed the removal and skipped the construction. It got the disruption without the strategy, and it paid for the privilege out of its own AI budget. That sentence is the whole of Part II compressed into one line, and it is the reason the productivity programme was never the problem — the missing half was.

Where this book stops

The moment the model says the unit must change, the question becomes what unit — and that is no longer a strategy question. It is a design problem: what the customer is promised, where the perimeter of that promise sits, how delivery stays adaptive inside it, and how the economics improve between engagements rather than within one.

One line is worth carrying across the seam, because it is the test that decides whether a candidate successor is real: what old commercial unit becomes smaller, what new unit becomes larger, and which valuable assets migrate from one to the other. And the sentence that gives it teeth — “a candidate that weakens nothing of the old economic unit is a good AI project. It is not the successor.”

Which is a hand-off, not a method taught here.

Worse, an AI-native entrant isn’t obliged to stop at “80 hours instead of 100”. It can ask why the customer is buying hours at all.

That question is the hinge between the case and the method. The rest of this book is about how to ask questions of that shape deliberately — before someone else asks them of you.

Two questions, both of which sound like serious strategic work:

• “What if AI adoption in our sector hits 100 per cent?”
• “What if competent generic advice is effectively free?”

One of them is worth a morning of the most expensive people in the firm. The other is worth nothing at all, and will produce a fluent discussion that ends with everyone agreeing. Telling them apart takes about thirty seconds once you know what to look for — and knowing what to look for is the whole content of this chapter.

Where the instrument comes from

The move itself is old and is not ours. Pushing one variable to an extreme until the geometry of the situation forces an answer is the pivot, and it is the human half of the partnership described in The Reshape. Its industry-scale form — Strip, Stretch, Stress, Stage — belongs to The Terminal Value Doctrine, and the observation that compression aims a search while adversarial search exhausts the region it aims at belongs to The Cognition Dimension Ladder.

None of them supplies a filter. They tell you how to run a boundary case; they do not tell you which boundary cases are worth running. Under the cheap-thinking paradox that gap matters, because the constraint is no longer the cost of running the search — it is the attention of the four people who could change a decision.

The pivot test

If pushing the variable to zero or to infinity collapses or transforms the set of legal moves, you have a pivot. If pushing the variable to either extreme just makes the situation “more of itself” — more expensive, slower, larger — but leaves the legal moves unchanged, you don’t.

Operationally, for a firm: at the extreme, does the set of things you are allowed to do change shape, or only change size?

Shape means new options become available and old ones become illegal — a revenue line that cannot be sold at any price, a structure that cannot be staffed, an offer that could not previously have existed. Size means the same options, scaled up or down. Bigger, cheaper, faster, poorer.

Six candidates, tested

Start with the two from the opening.

“AI adoption hits 100 per cent.” At the extreme, everybody has the tools. Your options are the same options, run faster, at lower cost, by everyone including your competitors. Nothing that was legal becomes illegal. Nothing that was impossible becomes possible that was not already possible at sixty per cent adoption. Fails — and the tell is that the answer at the extreme is “we’d all be busier”.

“Competent generic advice is effectively free.” At the extreme, an entire category of promise you currently sell becomes unsellable at any price. A pricing structure becomes indefensible. A hiring model — the pyramid, the graduate intake, the leverage ratio — turns from an engine into a liability. And something that is uneconomic today becomes viable: a standing advisory commitment, priced as continuity rather than as a project, which nobody can afford to staff while first-pass work is expensive. Options die and options appear. Passes.

Now four more that a real firm would actually consider — and the two failures teach more than the passes.

The failures are the useful part. “Our largest client fails” is a serious risk that deserves serious planning — it is simply not a strategic search question. It is a contingency, and contingencies belong in risk management with an owner, a trigger and a mitigation. Putting them in the search ledger is how a ledger fills up with twelve rows and produces no strategy.

Most of what fills a strategy offsite fails this test. That is not a criticism of the people in the room; it is what happens when a method has no filter and everything urgent gets treated as everything structural.

A probe is not a prediction

Say this every time the free-advice probe is used, because it is the sentence most likely to be quoted back without its context.

It is a probe: a deliberately extreme value chosen because the extreme makes the structure visible. This is why thought experiments work under Fog conditions when forecasts do not — they exchange timing certainty for structural certainty.

The consequence is worth stating in the form a chair can use. You don’t need to know whether that occurs in February 2027 or October 2028. If the answer is that a large share of your commercial proposition disappears when competent advice becomes free, you have learned something enormously valuable today. Note the shape of that finding rather than its size — the number is yours to compute, from your own last twenty engagements, and any figure printed here would be decoration.

Does the Fog ever lift?

No, and the honest version of that is stronger than modesty.

You don’t actually clear the AI Fog. You pierce it locally.

There is a mechanism behind it. A search apparatus that is good at its job manufactures Fog as a by-product of working well: every good search surfaces new candidate futures, and the solution space grows in proportion to how well you are searching it. We named that Permanent Fog, and framed it as something your own engine does to you.

Under the cheap-thinking paradox it generalises, and the generalisation is this book’s contribution to the idea: the recursion is not confined to your apparatus. Every other actor’s search is manufacturing futures you now have to hold — your customers’ searches, your competitors’, and those of firms that do not exist yet. Permanent Fog is not a property of your engine. It is a property of the market once cognition is cheap for everyone in it.

Which changes the posture in one practical way. Boundary cases are not a one-off stress test to be run when the strategy feels stale. They are a navigation instrument, run on a cadence, the way a chart gets refreshed rather than framed. And an instrument used on a cadence needs somewhere to write down what it found — which is the next chapter, and the reason it exists at all.

Who chooses the variable

The human, and this is not sentiment. Choosing which variable is live requires lived friction with the business — the specific knowledge of which assumption is load-bearing here, which comes from sitting with the situation rather than from having read about situations like it. Frontier models are very good at running a boundary case once you name the variable; they are noticeably less good at choosing which one to push.

What the machine adds is depth and stacking: exhausting the counterplay around a chosen extreme, and running the joint cases — two or three boundaries at once — that no human runs before lunch. The human supplies the pivot; the engine supplies the stack. That division of labour is described properly elsewhere in the corpus and does not need rebuilding here.

So: choose the variable with the pivot test, push it, and find out which assumptions fail. That is half a day’s work and it produces something real.

It also produces nothing at all, unless what it found gets written down in a form that closes.

The offsite went well. The boundary case was sharp, the discussion was the best the partnership has had in years, three people said “that changes how I think about this”, and someone photographed the whiteboard on the way out.

Six weeks later the budget is unchanged, the hiring plan is unchanged, the pricing sheet is unchanged, and nobody can reconstruct which futures were ruled out or why. The photograph is in somebody’s camera roll between a parking sign and a restaurant menu.

The insight was real. It was also, in the only sense that matters, free of consequence — and it will happen again next year, because nothing about the process was designed to prevent it.

Six columns

The rule that closes an entry

Without a decision delta, the exercise is scenario theatre — a wall of futures, a satisfied room, and an unchanged budget. Say it in those words inside the firm, because the phrase does more work than a policy would.

Two things follow, and they are what make the rule usable rather than moralising.

Open rows are honest rows. A ledger with three open entries is in better health than one with twelve closed ones, because the open ones are visibly waiting on something and everybody can see what. Openness is not failure; false closure is.

The rule also filters the front end. If you cannot imagine any answer to a boundary case changing a present decision, the case is not worth running — which is Chapter 9’s pivot test arriving from the other direction, and a useful cross-check on it.

How this relates to the Question Ledger

A reader who knows the corpus will already be asking, and pretending this is a new invention would be dishonest.

The Terminal Value Doctrine owns the Question Ledger: twelve fields, a gap audit, and the John West escalation from rejected answers up to the questions that generated them. It remains the definitional artefact and this book does not reproduce its schema. Its framing is the one to carry: strategy under classical conditions is a recommendation problem — pick the best option from a set. Strategy under AI Fog conditions is a search-quality problem — did the team actually search the question space, or did they converge on the first fluent answer?

The Strategic Search Ledger is that artefact’s six-column executive row, with one added field and one enforced rule. The division of labour is clean: the Question Ledger makes the search inspectable; this makes it consequential. A board can hold both — twelve fields where the search is run, six where it is governed — and should.

One completed row

The firm from Part II, three weeks after someone finally ran the arithmetic.

What each column bought

The rejected future is what makes the search real. Anyone can list futures; rejecting one costs something, because it removes an option the firm was quietly relying on. “Hold and absorb” was the partnership’s unstated default. It is now written down as rejected, with a reason, which means nobody can drift back into it without arguing.

The decision delta is what makes it strategy rather than analysis. Two practice areas and a hiring plan moved. If a row cannot point at something like that, the row is a memo.

The next question is what makes it a ledger rather than a document. It has a trigger, so the row can be reopened by evidence rather than by mood — and the trigger is observable by someone other than its author.

The absence of a date is deliberate and worth pointing at. Nothing in that row required knowing when anything happens. The firm changed its pricing and its hiring on the basis of a structural finding, and would have made the same decision whether the boundary arrives in eighteen months or five years.

Without a decision delta, it is scenario theatre: a wall of futures, a satisfied room, and an unchanged budget.

How many rows, how often

Three to five live rows is a working ledger. A firm running twenty is doing research, not governance, and will close none of them. Refresh on a cadence — the way a chart is refreshed rather than filed once, because under permanent Fog the search is a standing activity rather than a project.

And one change to the board’s standing agenda, which costs nothing and changes what gets prepared: at the start of a strategy discussion, ask where is the ledger, and which rows moved since last quarter?

One row, done properly, in a morning. The instrument works.

Which raises the question the method has been avoiding: is there anywhere you should not point it — any question where all this discipline is wasted effort, because cheap cognition is quietly making that question easier rather than harder?

Yes — and the domain where cheap cognition makes uncertainty fall is one every reader has sat through the results of.

Audit has spent its entire professional history sampling. Not because sampling was ever thought to be a good way of finding problems, but because examining everything was impossible at human cost. So the profession built an elaborate, defensible apparatus around a budget constraint: materiality thresholds, risk tiers, sampling schedules, statistical confidence intervals. All of it downstream of the fact that a person had to read each item.

That constraint is dissolving. KPMG describe AI moving fieldwork “beyond sample based testing by reviewing full data populations, identifying anomalies, trends, and exceptions, and detecting patterns that merit further investigation” — instead of testing a sample of transactions, analysing the full population¹⁰. Independent trend analysis reports the same shift and adds the second-order effect: it raises “regulator expectations of what is now technically achievable”¹¹. Vendors in the space claim analysis of 100 per cent of financial transactions across order-to-cash, procure-to-pay and record-to-report — a vendor claim, labelled as one, and not needed for the argument¹².

Why nothing multiplies here

Walk the difference rather than asserting it, because the difference is the chapter.

The population of transactions is finite and enumerable. You can count what must be examined before you start, and the count does not change because you got better at examining.

The test is well-defined and stable. A duplicate payment is a duplicate payment. It does not become something else because a competitor changed their pricing or a new entrant appeared in the market. Nobody else’s move redefines what you are looking for.

And more thinking converges. Each additional unit of cognition reduces the unexamined remainder. It does not create new categories of thing to examine.

Sampling was never a methodology anyone loved. It was a budget wearing a methodology’s clothes. Dissolve the budget and the uncertainty falls with it — which is exactly the promise the industry makes about AI, delivered in full, in this kind of question.

The boundedness test

Three conditions, in the same register as Chapter 5’s three — the two tests are siblings, and a firm should be able to run both on a whiteboard:

1. The population is enumerable. You can list the things to be examined: transactions, assets, tickets, contracts, claims, clauses, sites, invoices.
2. The evaluation function is stable. You know what a good answer looks like, and nobody else’s move redefines it.
3. The action set is closed. Finding more does not create new kinds of action; it changes which of a fixed set you take.

Now run strategy through it, and watch it fail all three. The population of futures is not enumerable — that was Chapter 3. The evaluation function is not stable, because what counts as a good strategy is defined partly by other people’s moves, which is what makes it strategy rather than optimisation. And the action set is emphatically not closed: every discovery creates new kinds of option, which is the recursion Chapter 9 described.

Three failures out of three. The same technology that clears an audit thickens a strategy, and it is not behaving inconsistently — the questions have different structure.

A second bounded case, in a different domain

Audit is a convenient example, so it should not carry the test alone.

Consider an asset-heavy operator — a utility, a network, a fleet. Condition reviews are run thoroughly on the assets already flagged as high-risk or due this cycle. Hundreds of others receive checklist-level attention. Nobody claims the unflagged assets are free of early-stage problems; only that expert hours cannot follow every transformer, corridor segment and contractor record with the same intensity. The sample becomes the world.

Our own Cognition Scarcity Audit puts the probe for that situation precisely: “If cognition were nearly free overnight, which population should receive a regenerated evidence-based argument — not just a score — that we currently touch only by sampling?”

Enumerable population. Stable evaluation — a degrading asset is degrading regardless of the market. Closed action set: inspect, repair, replace, defer, monitor. Bounded, three out of three. More cognition means less uncertainty, and there is no boundary case to run.

The audit that contains its own boundary

Here is the part that makes this chapter more than a hedge.

That same scarcity audit names four signatures of work an organisation does not do because attention was expensive. Run all four through the boundedness test and they do not score the same.

Three of the four are bounded. Cheap cognition simply fixes them, and firms should go and fix them this year without a boundary case in sight — every anomaly investigated, every asset argued, every contradiction between policy and practice surfaced without waiting for someone to nominate the question.

The third is unbounded, and it is precisely where more cognition produces more futures. The same audit contains the boundary of the paradox — which is a better argument than anything this book could have constructed for the purpose, because it was written for a different reason.

Note the honesty in row four. Cross-silo synthesis has a mostly stable evaluation function: what counts as a contradiction worth surfacing drifts with context, and a system that surfaces everything will bury its readers. The test is a diagnostic, not a sorting machine, and any row scoring “mostly” deserves a second look before you point a large budget at it.

So what do you actually do with two kinds of question?

Sort your questions before you spend cognition on them.

Flood the bounded ones. There is enormous, uncontroversial, unglamorous value sitting in census-instead-of-sample work, and none of it requires a strategy, a boundary case or a ledger. It requires a list and a budget.

Discipline the unbounded ones. That is where undirected thinking manufactures the Fog you are trying to see through, and where Chapter 9’s filter and Chapter 10’s artefact do their work.

There is a pleasing consequence here for the programme Chapter 5 criticised. Pointed at bounded questions, a productivity programme is straightforwardly good — and this is where its budget should go. What Chapter 5 objected to was never productivity itself; it was the pairing of productivity with a compressible commercial unit and nothing else. Sorted properly, the same programme stops being a depreciation accelerator and starts being what it was sold as.

Two tests now, and between them they cover the diagnosis: three conditions for whether your productivity programme is anti-strategic, and three for whether a question is bounded.

Which leaves the discipline itself. It can be adopted, understood, agreed with — and still fail, in ways specific enough to name.

The most likely failure is not the firm that never searched. It is the firm that adopted the artefact and hollowed it out — and that failure is worse, because the artefact is now providing cover.

Picture the ledger. Twelve rows. Every column filled. Confidence ratings on all of them, mostly high. Reviewed quarterly, presented to the board, admired by a non-executive director who says it is the best strategy documentation they have seen from a firm this size.

Then read the fifth column. Consider fixed-fee pilots. Explore alternative pricing. Monitor competitor moves. Continue to assess. Twelve rows of verbs that commit nobody to anything, in a governance artefact that now certifies the absence of a decision.

Six ways it goes wrong

Three of these are inherited — the Terminal Value Doctrine named the wrong-variable trap, forecasting in disguise and the fluent answer, and the Question Ledger chapter named empty fields and aspirational confidence. They are listed here with their tells rather than re-argued. The sixth is new, and it is the one this book is uniquely placed to name.

The sixth deserves its own paragraph, because it is the failure mode a good firm reaches. A search apparatus that works manufactures Fog as a by-product of working — the point Chapter 9 made about your own engine. Point cheap cognition at an unbounded question and it will happily produce candidate futures forever, each one plausible, each one requiring consideration, none of them closing anything. The ledger becomes a machine for generating the condition it was built to navigate.

So how much searching is enough?

Part III raised this and could not answer it. Here is the answer, and it is the one thing in this book that has no antecedent in the corpus.

Three consequences, because the rule is easy to state and easy to misapply.

It bounds a search that is otherwise unbounded. Coverage is not available as a stopping condition — the Fog does not lift, the space keeps growing, and a team that waits for completeness produces the twelve-row ledger with nothing in the fifth column. Decision-neutrality is available, and it is checkable in the room: if this future were true, what would we do differently? — nothing — then we’re done here.

It explains the artefact’s shape. Six columns rather than twelve, because the Strategic Search Ledger is a governance instrument and governance needs the decision rather than the whole search. The full search lives in the twelve-field Question Ledger, where the work is done. Two artefacts, two jobs — and the shorter one exists because a stopping rule needs something short enough to close.

It is a resource-allocation rule, not a modesty rule. Cognition is cheap and can be spent freely; that is the entire premise of the book. What is expensive is the attention of the four people who can actually change a decision. The stopping rule rations their time, not the machine’s. Let the apparatus run wide; close the row when the humans have what they need.

What a real gap looks like

The test is simple enough to apply in the meeting: can you name what evidence would fill it, and who would go and get it? If not, it is a feeling about the topic rather than a gap in the search.

The condition underneath all six

The failure modes are symptoms. The condition is that nobody owns the terminal-value question.

Efficiency reports to operations. Delivery reports to the practice leads. Pricing reports to the commercial director. Client relationships report to the partners. And the question of whether the unit survives — the one Part II spent three chapters on — reports to nobody. So the ledger becomes whoever’s side project it is, reviewed by people with day jobs that reward continuity, and side projects close rows with intentions.

The fix for all of it is one line, and it is the last thing this book has to say: a named owner, a cadence, and a board that asks for the ledger before it approves the strategy.

Which is a Monday-morning problem, not a philosophical one.

The first two are diagnostic and this book has answered them for one firm shape: generic cognition becomes cheap, and any commercial unit denominated in it stops being valuable at the rate the compression lands. The third is the asset question, and it belongs to the doctrine this book extends. The fourth hands off to a different book entirely, which is the point of the seam.

So: three moves, small enough to start this quarter, specified so nobody needs us in the room.

Move one — name the unit, test it for compressibility

Write down what the invoice is denominated in. Not what you sell — what the line item counts. Hours, days, seats, matters, tickets, reports, licences, transactions. Most executive teams have never written this down, because the invoice is treated as an accounting artefact rather than a strategic one.

Then ask what happens to that count when the work behind it takes forty per cent less time and the client can see that. Run the three conditions from Chapter 5 as a checklist: is the unit compressible; is the compression visible or performable; is entry above open.

If the answer is that revenue falls, your productivity programme is a depreciation accelerator. Keep it, re-scope it, stop calling it strategy. Move its budget toward the bounded questions from Chapter 11, where the same programme is straightforwardly good and its metrics mean what they appear to mean. Then take the strategic attention you have just freed — which is the scarce input, not the cognition — and spend it on move two.

Move two — run one boundary case, chosen with the pivot test

One variable. Pushed until the set of legal moves changes shape rather than size. Not five scenarios — one, worked until something breaks.

Who is in the room matters more than how long it takes: the people with lived friction in the business, because choosing which variable is live is the human half and it cannot be delegated to either a machine or an adviser. Half a day is enough.

The output is not a view of the future. It is a list of which of your assumptions are load-bearing, and which one just failed at the boundary. No date is required for that finding to be useful, and any conversation that turns to timing within the first ten minutes has drifted into forecasting.

Move three — close one ledger row with a decision that changed

Six columns. The fifth is not optional.

The test at the end of the exercise is blunt and takes ten seconds: if the budget, the hiring plan and the pricing sheet are all exactly what they were before, the search did not happen. You had a conversation about the future and filed it.

One row done properly beats twelve done as theatre, and a firm that cannot do one will not do twelve.

That last line is the cheapest change on the list and the one with the most leverage. A board that asks for the ledger before approving a strategy gets a different pack prepared for it, by different people, containing evidence of a search rather than a well-argued recommendation.

Do we possess a machinery for directing abundant cognition at the questions that determine whether this company still deserves to exist?

Scott’s own version of that question is less polished and lands harder: in a world with cheap thinking you have to ask what your business means, what your new value is going forward, what the terminal value of your company is, and how you pierce that Fog. It is an uncomfortable thing to put on the agenda of a firm that is currently profitable. That discomfort is the whole reason it belongs there, and why it will not appear unless someone with authority puts it there deliberately.

Where this book ends

Once the ledger says the unit must change, the question stops being strategic and becomes a design problem: what the customer is promised, where the perimeter of that promise sits, how delivery stays adaptive inside it, and how the economics improve between engagements. That is the subject of AI-Native Service Architecture, and this book stops at its door. The bridge across is one test: what old commercial unit becomes smaller, what new unit becomes larger, and which valuable assets migrate.

What this book contributed

Boundary-case reasoning is the discovery mechanism of the Terminal Value Doctrine. Car Discovery is its organisational form. The AI-native successor offer is one of its commercial outputs. And underneath all three is cheap cognition.

This book owns the first of those three, and it added three things to it. A causal reason the mechanism is necessary at all — the Fog is partly endogenous, and cheap cognition is one of the things thickening it. An arithmetic showing what happens to a firm that ignores that and optimises anyway. And a closing rule that turns a search into a decision rather than a document.

Which is a modest contribution stated precisely, and it is why a series of apparently adjacent subjects — thinking, thought experiments, cognition economics, the Fog, terminal value, successor offers — increasingly looks like one theory of the firm under abundant cognition rather than several.

What none of this will do

It will not clear the Fog, and the book should not end by implying otherwise.

The same cheap cognition that lets you search is funding everyone else’s search. A good apparatus manufactures new possibilities as a by-product of working well. The solution space will be larger next year than it is now, and larger again the year after, and there is no version of this discipline that reverses that.

What the discipline does is trade timing certainty, which you cannot have, for structural certainty, which you can — and leave behind a record of which futures you rejected and why. So that when one of them starts arriving, the firm is executing rather than debating, and the argument about whether it was coming has already been had and written down.

AI makes thinking cheap, but it does not make strategy easy. Cheap thinking destroys old scarcities while multiplying the number of plausible futures. The board’s job therefore shifts from extrapolating yesterday to allocating disciplined cognition against terminal-value questions: identify the load-bearing variable, push it to the boundary, search what survives, and build the successor before the market does.

The firm that goes twenty per cent faster at selling hours has not answered any of that.

It has arrived at the same place sooner.