The Re-Roll: Why AI Reprices Every Company's Character Sheet at Once — and Strategy Moves to the Industry

Imagine the world as a long-running game of Dungeons & Dragons. Every company at the table has a character sheet. The attributes are familiar: scale, brand, capital, distribution, data, licences, relationships, systems, people. Everyone knows the incumbents. Everyone knows everyone's strengths and weaknesses. Everyone knows how battles are fought, because the same battles have been fought, with minor variations, for decades.

Now imagine that between one session and the next, every character sheet at the table is re-rolled. Not yours. Not your rival's. Everyone's. The stats that made the strongest player strong have been re-priced. Some items in the inventory turned out to be more valuable than anyone thought. Some turned to lead overnight. And nobody at the table has finished re-reading their own sheet — let alone anyone else's.

That is the strategic condition AI has created, and most strategy work has not caught up with it. AI did not add a new weapon to the old game. It changed the rules, redrew the map and re-rolled every character sheet. Everything is up for grabs again — not because history no longer matters, but because history has been repriced.

This article makes one argument by one route: because every firm's inherited attributes are repriced simultaneously, the unit of strategy has moved from the firm to the moving value system around it — and capital and time must follow the value. Along the way it explains the most confusing pattern in the current market: why a company can genuinely lead its sector in AI adoption and still watch its position erode.

1. The re-roll: your character sheet has already been rewritten

For decades, companies competed with largely understood attributes. Business models changed, but slowly enough that a firm could plan, defend and compound within a recognisable set of rules. An incumbent's advantages were legible: the balance sheet, the brand, the distribution agreements, the customer relationships, the regulatory licence, the decade of operational data.

Then cognition became abundant. Software became generative. Expertise became callable. Coordination became cheaper. Products that once required departments could be built by small teams. Customers acquired their own intelligence. Competitors became cheaper to create.

The scale of the input-price collapse is worth stating precisely, because it is the physical cause of everything that follows. Epoch AI measured the cost of buying a fixed level of model performance over time and found that the price of matching GPT-4 on PhD-level science questions fell by roughly 40× per year — with declines ranging from 9× to 900× per year across the benchmarks they measured.¹ Gartner projects that inference on a trillion-parameter model will cost over 90 per cent less in 2030 than it did in 2025.² When an input to nearly every business model gets cheaper at that rate, every attribute that monetised the old scarcity gets a new modifier.

The consequences do not land evenly, and this is the part boards most often miss. The re-roll is not a uniform devaluation of the past. It is a repricing, and repricings have winners on both sides of the ledger:

• Some inherited assets become more powerful. A regulated licence, a trusted brand, physical capacity, real distribution — attached to abundant cognition, these can appreciate.
• Some can travel into the future — but only if they are converted. Decades of engagement history, claims data, methods and playbooks contain latent value, but unconverted history is not a moat. It is memory that eventually walks out the door.
• Some become stranded. Assets whose returns depend on a scarcity AI is dissolving — the analyst pyramid, the reporting factory, the per-seat licence ladder.
• Some old advantages become liabilities. The customised platform that locked customers in becomes the cost they resent; the process empire becomes overhead.
• Some liabilities become raw material. The unloved legacy system becomes a behavioural specification an AI can observe, test and regenerate.
• Some moats become maps — showing an attacker exactly what to rebuild, minus your cost base.

The empirical signature of a genuine re-roll — as opposed to another technology cycle the incumbents absorb — is that entrenched advantages stop translating. Menlo Ventures' enterprise data found exactly that at the AI application layer: startups captured 63 per cent of the market, up from 36 per cent the year before, earning nearly two dollars for every incumbent dollar. Their own commentary makes the point better than any theory could: "On paper, this shouldn't be happening. Incumbents have entrenched distribution, data moats, deep enterprise relationships, scaled sales teams, and massive balance sheets."³ On paper — meaning, on the old character sheet. The sheet has been re-rolled.

The past still matters — but it no longer carries its old score automatically.

And this is happening to every company, in every industry, at the same time. The simultaneity is the strategic fact. When one player's sheet changes, that player adapts. When every sheet changes at once, the board itself is in motion — and firm-level analysis, however rigorous, is analysis of the wrong object.

2. The unit of change is now the industry

Most strategy still looks at one company at a time. How should our company use AI? How should our people become more productive? Which of our workflows should be automated?

AI is operating one altitude higher. An industry is a bundle — of capabilities, scarcities, institutions, agreements and customer behaviours, held together by the transaction costs and cognitive costs of its era. AI unbundles that bundle, reprices every component, and makes the pieces available for recombination. Watch what the pieces become:

None of these is hypothetical. Customers are already becoming producers of workflow software: Retool's build-versus-buy data found 35 per cent of teams have already replaced at least one SaaS tool with a custom build, and 78 per cent expect to build more custom internal tools in 2026.⁴ The category isn't losing a competitor fight; it is being partially unbundled by its own customers.

The old unit of strategy was the firm. The new unit is the moving value system around the firm. That changes the question a board should be asking. Not: how do we make this company better? But:

When this industry is recombined, where will value live, who will control it, and what must we become to arrive there first?

Every clause of that question does work. Where will value live — because value moves rather than disappears. Who will control it — because the entity that collects the value may not yet exist. What must we become — because arriving first may require constructing something your current organisation would not naturally build. This is the industry-scale extension of the Terminal Value Doctrine's board method — pushing a structural variable to its boundary and asking what still deserves to exist — applied not to one firm's portfolio but to the whole table.⁵

3. Why adoption leaders still lose: the compounding asymmetry

Here is the pattern that breaks the standard mental model. A company rolls out AI aggressively. Copilots deployed, pilots running, adoption dashboards green. By every internal measure the programme is a success. And the firm's competitive position does not improve — sometimes it visibly erodes. The instinctive diagnosis is execution failure: wrong tools, wrong training, wrong change management. The actual explanation is structural, and it has nothing to do with execution quality.

AI adoption is an operating question. AI pressure is a strategy question. Adoption is something your firm does inside its walls. Pressure is something the market does to your industry — and it arrives through three channels at once, because the same price collapse that funded your pilots funded everyone else's moves too:

Customers use AI to perform work they once purchased. They do not need to change suppliers to disintermediate an incumbent. They can simply buy less. The demand side is not waiting politely: in legal services, research collected by Clio found 67 per cent of corporate legal departments expect AI to change how hours are billed, 71 per cent of buyers already prefer flat fees — and, in the line that should be read aloud in every partner meeting, "clients have moved faster than the industry has."⁶ The demand side repriced before the supply side decided to.

Competitors do not merely become more efficient. They become cheaper at changing what they are. Designing an offer, modelling its economics, testing it against a segment — the cost of attempting structural change has collapsed, so the rate of attempts rises. Public markets have started pricing the consequence for incumbents: Sapphire Ventures' software indices fell 20 per cent — pure SaaS 23 per cent — through mid-February 2026, decoupling from an essentially flat Nasdaq, with the authors naming AI displacement risk directly: "risk is up, and terminal value assumptions are down."⁷

Constructors no longer need the capital, headcount or permission once required to assemble a credible alternative. Venture funding is concentrating rather than scattering — Crunchbase recorded $300 billion invested in Q1 2026 with AI taking 80 per cent, while seed deal counts actually fell 30 per cent⁸ — but the constructor channel was never mainly about funded startups. It is about the falling cost of construction available to everyone: the customer who builds instead of buys, the mid-sized rival who ships what used to need a platform team, the two employees who leave with a clear view of one broken workflow. And it may be you.

Now put the two clocks side by side and the adoption-leader paradox dissolves:

Its cognition compounds inside the walls at the speed of its change programme. Everyone else's cognition compounds across the market at once.

An internal AI programme compounds at the speed of your training calendar, your governance committee, your rollout schedule. Market pressure compounds at the speed of every customer, competitor and constructor discovering, in parallel and without coordination, what cheap cognition lets them do. Those are different exponents. Internal excellence does not close that gap, because productivity applied to a depreciating commercial unit simply produces the old value faster. If the unit you sell is being repriced, being faster at producing it is not a defence — it is a more efficient way of harvesting a declining asset without noticing.

The gap between activity and result is now measured at market scale: BCG finds only about 5 per cent of companies are achieving substantial financial gains from AI while roughly 60 per cent see little or no material return despite heavy investment — and the 5 per cent show roughly four times the three-year total shareholder returns of the laggards.⁹ The differentiator in that data is not spend, and it is not adoption rate. It is altitude: the winners are re-positioning against where value is going, not accelerating what they already do. The task is not to accelerate whatever already exists. The task is to discover what still deserves to exist.

4. Value does not disappear — it moves, and it moves at a speed

AI does not only automate work. It can remove the reason the work existed. When a scarcity collapses, the value that scarcity supported does not evaporate — it migrates, and it does not wait for the incumbent to approve the destination. Generic analysis, generic software, generic content, generic advice: all becoming cheaper. Value moves towards what remains difficult to reproduce — judgement, trust, taste, evidence, accountability, authority, proprietary context, relationships, the willingness to carry consequences.¹⁰

Destruction at one layer is creation at another. Compression in one part of an industry creates expansion somewhere else. An old bundle may disappear while a new set of specialised transactions multiplies around it. The question is never whether there will be value in your industry; it is which layer it will settle in, and who will be standing there.

This movement has both a direction and a speed, and strategy needs both:

Direction without time is not strategy. Arriving at the right destination after the opportunity has closed has no economic value. And the time dimension has its own arithmetic, because two clocks are running simultaneously. The existing model has a runway — how long the old economics keep producing cash. The successor has a proof clock — how long before it must produce real evidence, while the existing model can still fund it. The market's branching rate keeps generating new possible futures whether or not you respond; a firm that waits for clarity is not pausing the game, it is spending runway without starting the proof clock.¹¹

Delay is the only move that worsens both clocks at once.

This is why urgency in the AI era is not hype — it is arithmetic. The market does not preserve an opportunity because an incumbent needs another quarter to become comfortable.

5. Capital must move with value

Capital allocation is strategy made visible. Every dollar, every hire, every acquisition, every product roadmap and every quarter of executive attention is a vote for a particular future. Under re-rolled conditions, the first act of honest strategy is to reclassify your assets by how they behave as AI improves — the Terminal Value Doctrine's three classes: stranded assets, whose returns depend on a scarcity AI is dissolving; convertible assets, whose history, data, methods and relationships can travel into the future but only if converted into portable, machine-usable form; and compounding assets — proprietary context, evaluation systems, evidence, governance, customer intelligence — which every model release makes more valuable.⁵

The capital response follows directly, at asset level and at company level:

The old business may still produce excellent cash. Harvesting it is not betrayal; it is how the transition is funded. But operational excellence buys runway. It does not choose the destination. A board that runs only the harvest motion has decided — usually without saying so — to fund somebody else's construction.

Standing on the sidelines is still a capital-allocation decision. It allocates the future to somebody else.

6. You will be recombined — the only question is your role

No company gets to choose whether its industry will be recombined. It chooses only the role it will play: raw material, dismantled and repriced by other people's recombinations; spectator, watching customers, competitors and constructors unbundle its value proposition around it; or author, using its capital, customers, trust, data, distribution and domain knowledge to build the recombination itself.

The authored path has a name: self-disintermediation. Ask how an AI-native attacker would rebuild your industry — then build that attacker yourself, before someone else does. Build the product that makes your current product less necessary. Build the company that attacks your current economics. Build the business model your existing incentives would prefer not to imagine. Do it while the old model can still finance the new one.¹²

Incumbents rarely do this naturally, and the reason is structural rather than stupid: the org chart is built to scale yesterday's decisions, and no product team volunteers to design its own replacement. That is why the half-measures fail. Bolting AI features onto the legacy model produces the numbers we keep seeing — Activant Capital reports fewer than 4 per cent of Salesforce customers paying for Agentforce, the incumbent's own AI agent product, inside its own installed base.¹³ An AI feature on an old model is not a successor. It is a faster horse with a chatbot.

Incumbency, in other words, is not automatically an advantage. It becomes an advantage only when the incumbent's assets are attached to the future value layer rather than used to defend the old one. History matters only when it can travel.

7. AI-native is an economic test, not a tool count

Which raises the definitional question: what would it mean to have actually made the transition? Not Copilot seats. Not pilots, committees, announcements, or the percentage of code written by a model. Those are inputs, and inputs can be theatre.

The test runs at every frontier release. Ask one question when the next major model arrives: did our company just become stronger — or more exposed? If improved AI lets customers replace more of what you sell, compresses your margins and weakens your differentiation, you are still operating the legacy model, whatever your AI budget says. If improved AI makes your proprietary context more useful, your delivery machinery stronger, your evidence systems more powerful and your products more valuable, the sign has flipped: the same force that is destroying the old model's economics is now compounding yours.¹⁴

The reason to define AI-native this way is that a derivative cannot be performed. Adoption metrics can be gamed; a sign flip cannot. And the proof must appear in behaviour: paid demand, customer budget moving into the new unit, repeatable delivery, capability that compounds, a successor that can win without depending on the same hero every time. Activity is not evidence. A polished future is not a proved future.

8. What this looks like when someone runs it

Doctrine earns its keep in contact with reality, so it is fair to ask what operating on these assumptions actually looks like. I'll use the venture engine my co-founder and I run — SongbirdX, which builds AI-native companies — not as a pitch but as a specimen, because it was built by taking the re-roll literally.

The engine works at the level where industries are being unbundled, repriced and recombined: identify the scarcity that is disappearing, the scarcity that is emerging, the value layer that is moving, the assets that can cross, and the combinations that have only now become possible. Then turn that signal into an AI-native company, product or business model — thesis, model, brand, product, market — with a first version in the world in 30 days.

Thirty days is not bravado, and it is not the time needed to invent the underlying thought. Most of the cognition has already been capitalised — years of frameworks, methods and judgement compiled into forms the engine can call on. Thirty days is the compile and contact-with-reality period: the deadline that forces the first coherent venture to meet the market before the insight that created it goes cold and before the market branches again. What a proof cycle must produce is deliberately unimpressive-sounding: a proposition in the world; a customer response; a live product surface; an answer that could disappoint us. That last item is the point. Build options, put them into the world, let reality eliminate the weak ones, and allocate serious capital only when evidence earns it. This is not innovation theatre. It is capital allocation made executable — and the same loop works inside an incumbent constructing its successor as it does inside a venture studio.

The declaration

Doctrines bind behaviour or they are moods. The operating commitments of this one are short enough to state in full. We will not automate what should no longer exist. We will not confuse more ideas with better strategy. We will not mistake productivity for transformation. We will not defend stranded assets until they consume the capital required to escape them. We will not call a faster legacy model AI-native. We will not wait for certainty that can arrive only after the opportunity has closed.

And the affirmative side: search wider; choose harder; convert history into portable capability; build the attacker before the attacker arrives; force consequential uncertainties to meet reality; allocate capital towards what should exist next.

The world has been re-rolled. Everything is up for grabs — not because history no longer matters, but because history has been repriced. Every industry will be recombined. Every company will participate. The only choice is whether you help write the new combination or are written out of it.