We Were Never in the Same Conversation

How AI strategy stalls before it starts — and how to reach the minimum shared premise.

Scott Farrell · LeverageAI · August 2026

You leave the room thinking you had a strategy conversation. You used the same words: transformation, innovation, AI roadmap, productivity, competitive advantage. Everyone nodded. Then the capital went to copilots, a chatbot, and a pilot that fits inside last year's process map.

That is not a failure of explanation alone. It is often evidence that you were never in the same conversation.

When I look back at a management degree completed years ago, I used to believe the value was skills — how to run a company, how to read a value chain. Some of that was real. But a large share of the value was quieter: a shared language. Terms such as opportunity cost, organisational design, and competitive advantage were not decoration. They were cognitive compression. Two people who shared them could move from "this activity costs a lot and customers do not seem to value it" to "this part of the value chain may not be strategically differentiating" without rebuilding the whole model in the room.

That is an observation, not a study. It is also the shape of the current stall. Practitioners who have lived inside AI systems for years — built with them, watched them fail, watched software and cognition reprice — have effectively done an MBA in AI. Most business leaders have experienced a fraction of that. They may know the vocabulary. They do not share the world-model.

So when you say "AI makes cognition cheap," you hear custom software, workflow recomposition, stranded configuration, successor offers, judgment and trust becoming relatively more valuable. They may hear "ChatGPT can write emails faster." That is not a small difference in technical knowledge. It is a difference in economic world.

The AI strategy gap is not primarily that executives know less about AI. It is that the practitioner and the executive are reasoning from different economic worlds. The first job is not education or persuasion. It is to make those worlds explicit.

Three gaps, not one skills gap

What looks like one "AI skills gap" is usually three distinct failures stacked on top of each other.

1. The language gap

You and the executive do not yet share a dialect. A technically correct answer delivered in the wrong organisational register is operationally wrong. The executive needs the idea in the language of enterprise value, customers, strategic exposure, capital allocation and control — not models, agents, context windows and token economics. Our work on Register names this problem precisely: the enterprise translation prize is not language pairs; it is organisational dialect.¹

Language gaps are real. They are also the gap people most enjoy fixing, because training, glossaries and "AI briefings for the board" feel productive. Fix only the language gap and you will still stall if the premises differ.

2. The premise gap

This is deeper. You may share the words and still not share the starting assumptions.

You are reasoning from premises such as: useful cognition is becoming radically cheaper; software construction is becoming radically cheaper; capability change may be discontinuous; AI will mediate more customer and organisational activity; some existing sources of value will disappear; judgment, context, evidence, authority and trust become relatively more valuable.

The business leader may still be reasoning from: AI is another software feature; adoption proceeds at normal enterprise speed; the current business model is basically stable; AI should be evaluated through productivity ROI; the company can respond after the market becomes clearer.

You cannot reason together until the premise difference is visible. This is why AI Fog matters: the forecast horizon is shrinking while the number of plausible business architectures is expanding. Under those conditions, maintaining conventional forecasts as the shared method is not enough.²

3. The action-identity gap

Even after someone understands and provisionally accepts the premise, they may be structurally unable to act on it. Authority, compensation, professional identity and operating metrics are attached to the present company. The horse division is not naturally going to invent the car. Its job is to operate the horse business.

That is why apparently intelligent leaders can understand the argument and still retreat to Copilot rollouts and process automation. Those actions fit existing organisational machinery. Reconsidering what the company should sell does not.

A stalled conversation, diagnosed

Here is the shape that repeats. A practitioner sits with a senior leader — sometimes a partner, sometimes a business-unit head — to talk "AI strategy." The leader likes the word innovate. The practitioner hears an assumption: there is an existing process; we will make it better; the process, owner, customer and unit of sale all survive.

The practitioner tries to say something larger: large parts of the economy are repricing; even if your industry feels only second-hand impact now, demand and supply can move against you together; incremental steps will not answer a nonlinear bench problem. The leader agrees that "AI is important" and asks for an innovation roadmap. The roadmap becomes a list of tools inside the current commercial unit.

Apply the three gaps.

The unblock was not a better model demo. It was minimum viable premise: stop asking whether they "believe in AI," and ask which assumptions about the company's future they are no longer prepared to treat as safe. Once the room could name premises as risk rather than as faith, "innovate" had to resolve into a verb with portfolio meaning — and some of the proposed tools reclassified themselves as operating-tail work, not strategy centre.

Minimum viable premise is risk logic, not conversion

"Do you believe in AI?" is nearly right and slightly too binary. There is a quasi-religious element because people are reasoning from different beliefs about an unseen future. But a board does not need to become a congregation.

The more defensible threshold is:

Which assumptions about the future of this company are you no longer prepared to treat as safe?

That moves the conversation from faith to risk and capital allocation. A board does not need to forecast that AI will soon perform nearly all daily work. It only needs to accept three things:

1. There is a material probability that cognition and software become dramatically cheaper within its investment horizon.
2. The consequences for some current assets and profit pools would be severe.
3. Preserving an option against that future is therefore rational.

That is minimum viable premise — sometimes called minimum viable belief, but "premise" is cleaner because it refuses the conversion frame. The leader does not need to agree with your exact forecast. They need to agree that the current assumptions are no longer safe enough to remain untested.

Why the premises are no longer automatically safe

You do not need a vendor roadmap to open a premise conversation. You need public, named evidence that inherited defaults are weaker than they were two planning cycles ago.

The International Monetary Fund's June 2026 note on AI and cybersecurity in the financial sector frames the issue as financial stability, not IT fashion. AI is reshaping cyber risk by accelerating the speed, frequency and breadth of vulnerability discovery and potential exploitation. The core concern is less "new attack types" than scale effects across common technologies — shared infrastructure, common software, concentrated service providers — that can turn operational weaknesses into systemic events. The note emphasises controls that limit blast radius, robust recovery, and coordination that matches the cross-sector nature of the risk.³

The Australian Signals Directorate's April–May 2026 updates are more measured and still strategically important. Independent evaluation found that a frontier model was not dramatically more capable than prior models on individual cyber tasks, but could autonomously chain those tasks into an end-to-end intrusion — a meaningful uplift in overall capability. ASD also notes that many vulnerability-discovery techniques can already be reproduced with inexpensive open-weight models, so the assumption that hostile actors will lag frontier capability by many months is no longer safe. Existing defensive controls still create real friction when properly implemented; the point is not panic. The point is that "we can wait until this is clearer" is a weaker default premise than it was.⁴

Public assessments of a frontier lab's cyber-capable model sit in the same premise pack as incidental evidence: capability and diffusion move, restricted-access buffers erode, and waiting is not a neutral pause. Do not turn this into product marketing. Turn it into risk logic for the board.

"Innovate" is a conservation word

In ordinary business use, innovate usually means: keep the existing process, product, department, commercial unit and assumptions — but make something within them newer or better. That is why everybody likes it. Nobody has to concede that their process may be unnecessary, their product may be losing relevance, or their business model may be approaching expiry.

When someone says "How can we innovate this process?" the sentence has already decided that there should still be a process; that this is roughly the right process; that its objective remains valid; that its current owner probably remains the owner; that the customer will continue buying roughly the same thing; and that progress will come from improving the existing machine.

A better question separates outcome from process:

If this outcome still mattered, but we were starting today with AI, what would we build — and which parts of the current process would never exist?

Stop Automating, Start Replacing already named the binary: automation asks which cogs to improve; reimagination asks why there are fourteen cogs at all.⁵ This piece sharpens that binary into a boardroom instrument — a ladder of verbs that refuses to let improvement and replacement share one reassuring word.

The six-verb ladder

Innovate is too ambiguous to approve capital against without one of the above attached.

The utilisation paradox — worked arithmetic

Consulting shows why small erosions can become nonlinear — and why incremental framing is economically wrong, not merely aesthetically timid.

Damage need not arrive as one dramatic competitor replacing the whole firm. It arrives through simultaneous compressions:

1. Clients perform more first-pass cognition themselves.
2. Staff produce the same artefacts faster.
3. The market learns that the work requires fewer hours.
4. Competitors pass through some of the productivity gain in price or scope.
5. Demand and supply move against the firm at once.
6. Bench utilisation worsens — because slack absorbs the shock first.

A consultancy does not need to lose 30% of its revenue for the economics to break.

That is the whole proof. A modest demand reduction produces a disproportionate increase in bench.

Internal AI productivity can make the problem worse. If a project that previously required ten consultant-days now requires seven, the firm has not automatically created 30% more profit. It has created three more available consultant-days that must be sold into a market whose demand is also compressing.

AI makes each consultant capable of supplying more work at the same moment the market needs fewer consultant-hours.

That is the utilisation paradox. Our work on the postures of an AI-native consultancy already notes the trap inside "AI-enabled delivery": utilisation optics improve while the firm still sells the old broad, labour-priced unit.⁶ Faster decks are not a strategy for a market that needs fewer deck-hours.

So the response cannot be "innovate consulting" as an unqualified phrase. That phrasing permits proposal automation, chatbot ornaments and utilisation dashboards — all rational operational work, none of which answers: what will the customer buy when they no longer need the same quantity of consultant cognition?

The progression this book owns

The executive progression is:

Recognition → Language → Reasoning → Migration

• Recognition: Something structural has changed; old assumptions are unsafe.
• Language: Shared terms for value migration, asset classes, AI Fog, governance and successor construction — and verbs that force portfolio classification.
• Reasoning: Push variables to boundary cases and inspect a structured question set rather than demand confident forecasts. That method is the Terminal Value Doctrine's territory; this piece only gets you to the door.²
• Migration: Harvest, migrate and construct without requiring certainty or reckless commitment — named here as destination, not taught as mechanics.

Strategic Premise Alignment is the missing layer immediately before the doctrine. Terminal Value Doctrine names the altitude and the reasoning. Three-Lens aligns success definitions, error tolerances and measurement at build time.⁷ Neither addresses the epistemic precondition: participants inhabit different economic worlds and must make those premises explicit first.

What to do Monday

1. Diagnose the last stalled conversation with the three gaps. Write which one was load-bearing. If you wrote "skills," force yourself to choose again.
2. Rewrite the opening question from "Do you believe in AI?" to "Which assumptions about this company's future are you no longer prepared to treat as safe?"
3. Run the verb challenge on the current AI pack. Every "innovate" gets one of six verbs. Separate operating-tail items from portfolio-centre items.
4. Work the arithmetic that fits your unit of sale — consulting bench, product seats, project margin, support load. Show nonlinear slack, not a vague "disruption" slide.
5. Hand the public-evidence pack (IMF, ASD, generalised frontier cyber assessments) as risk context, not as conversion literature.
6. Only then open Terminal Value Doctrine altitude questions and, later, Three-Lens build-time alignment.

You were never trying to win a debate about whether AI is impressive. You were trying to get into the same conversation about economic reality — so strategy could begin at all.

When that happens, it feels less like persuasion and more like relief: same room, same premises, different futures finally discussable as capital rather than as faith.