AI is not only asking whether the existing company can move faster. It is asking whether the existing company is still the right shape.
That is the part of the conversation I think many leaders are still avoiding. It is easier to talk about AI use cases than to ask what those use cases imply. It is easier to automate a step than to question why the step exists. It is easier to add a generative agent to procurement, sales, marketing, product, or consulting than to ask whether the whole value chain was designed around constraints that no longer hold.
This is why I keep coming back to one uncomfortable question:
If cognition is no longer scarce, what parts of the firm were built around scarcity that no longer exists?
That question is not anti-management. It is not anti-process. It is not a fantasy about AI replacing the company. It is a practical operating-model question.
Most organizations were designed around human limitation. Expertise was scarce. Analysis took time. Synthesis was expensive. Customer insight had to be gathered manually. Strategy had to be packaged into decks. Leadership attention had to be protected. Decisions had to be batched. Governance had to slow things down because the cost of moving with incomplete information was high.
Some of that logic still matters. Humans still need judgment, observation, accountability, courage, and taste. AI does not magically understand consequences. It does not carry fear in the way a responsible human does. It does not know what it means to be accountable when a decision changes a product, a client relationship, a team, a budget, or a career.
But AI changes the cost of knowing.
And once the cost of knowing changes, the old shape of the firm deserves to be questioned.
The Shape Of The Firm Is The Real AI Question
A lot of AI adoption still happens at the surface of the organization.
Sales teams use AI to generate more leads. Marketing teams use AI to write more SEO content. Product teams use AI to summarize interviews. Procurement teams use AI to validate invoices. Consultants use AI to write decks faster. Managers use AI to produce meeting summaries and task lists.
Many of these use cases are useful. Some of them are genuinely valuable. I am not interested in dismissing them because they are not radical enough.
But they are strategically insufficient.
A use case improves one step inside the old value chain. Transformation asks whether the value chain still makes sense.
That distinction matters because companies can spend years becoming more efficient at the wrong level. They can add AI to broken workflows. They can accelerate bad decisions. They can produce cleaner artifacts for strategies that should have been challenged earlier. They can reduce the cost of a task while leaving the surrounding operating model untouched.
This is the band-aid problem.
A band-aid AI use case says: we have a process, let us make one step faster.
A serious AI operating model asks: if this capability existed from the beginning, would we design the process this way at all?
That is where the conversation becomes political, organizational, and economic. AI adoption is not just a technology program. It challenges the current shape of the firm because it challenges the machinery that protects the old value chain.
The old machinery includes budgeting cycles, role boundaries, approval gates, reporting rituals, steering committees, procurement logic, KPIs, managerial identity, and the consulting economics of selling certainty before evidence is strong.
AI does not make those things disappear.
It makes them visible.
I wrote about the lived version of this in The Old Workflow Is Too Small for AI. The MICRO problem is that individual work accelerates while the workflow stays slow. The MACRO problem is larger: many firms are still optimizing value chains built around cognitive scarcity.
That is a different level of ambition.
Design Thinking Is A Useful Analogy, Not The Whole Story
Design Thinking is a good place to see the difference between a band-aid and redesign.
The band-aid version is obvious. Use AI to replace some exploratory interviews. Ask a model to simulate personas. Summarize patterns. Generate themes. Produce insight clusters. Draft opportunity areas.
That may be useful. But if that is all we do, we have not transformed the process. We have only replaced one chore inside the old process.
The more interesting question is what the process becomes when human limitation is no longer the same constraint.
Traditional Design Thinking and Business Modeling processes often stretch across weeks or months because the work has to move through human bottlenecks. Research has to be gathered. Personas have to be defined. Interviews have to be conducted. Themes have to be synthesized. Workshops have to be scheduled. Concepts have to be drafted. Prototypes have to be made. Feedback has to be gathered. Decisions have to be prepared.
Some of that time is real value. Much of it is friction.
AI changes the balance.
If a team can use AI to explore a much larger evidence base before the first workshop, generate provisional persona hypotheses, test contradictions, pressure-test market assumptions, and create first-pass business-model alternatives, then the human workshop does not need to spend its energy on the old chores.
The humans can spend more of the week where humans matter most: observing, challenging, ideating, making tradeoffs, prototyping, testing, and deciding what the evidence actually means.
That is a different process.
The point is not to put AI in charge of ideation. I would not. AI can be useful in ideation, but I do not think it is naturally superior to humans at creative judgment, contextual observation, or productive discomfort. The point is to feed the human process with more evidence than it could previously absorb.
AI can increase the mass of evidence. It can generate incremental findings across personas. It can help a team ask sharper follow-up questions. It can turn the first answer into the beginning of a conversation rather than the destination.
But there is a difference between asking AI to generate likely interview results across personas and watching a 74-year-old who has never used a smartphone try to use a prototype.
The first can expand the evidence base. The second can validate whether the hypothesis survives contact with reality.
AI can help a team see more patterns before the room. It cannot fully replace the human act of observing hesitation, confusion, pride, embarrassment, workaround behavior, or the small physical details that reveal whether a product actually makes sense to the person using it.
This is the lesson for the firm. The AI opportunity is not simply faster research. It is a redesigned decision environment where humans spend less energy gathering and formatting and more energy observing, judging, and validating.
A three-month process does not become strategic just because it takes three months.
Sometimes it is just a process designed around old limitations.
Product Development Shows The Trap
Product development is probably the best example because the trap is so familiar.
The band-aid version of AI in Product is straightforward. Summarize 50 customer interviews. Draft PRDs. Organize messy notes. Generate user stories. Turn feature ideas into backlog items. Create variants of documentation. Scan support tickets. Extract themes from sales calls.
Again, all useful.
But none of that fixes the deeper problem if the organization is already good at building the wrong things.
Many product organizations do not fail because they lack speed. They fail because the wrong decisions were made before anyone opened the roadmap, the backlog, or Jira. They fail because executive ideas become commitments too early. They fail because every customer request is treated as strategy. They fail because shipping is rewarded more visibly than learning. They fail because evidence is gathered after the direction has already become politically difficult to change.
AI can make that worse.
If the operating model rewards feature throughput, AI will help create more feature throughput. If the roadmap is already a mechanism for laundering executive assumptions into product commitments, AI can make the artifacts cleaner and faster. If customer signal already arrives as filtered anecdotes, AI can make those anecdotes more presentable without making them more true.
That is the danger of use-case thinking.
It asks: how can AI help Product move faster?
The better question is: how can AI reduce the risk of building the wrong thing?
That is a value-chain question.
Product is not only about creativity. It is also about increasing the mass of evidence until the risk of a decision becomes clearer. AI is powerful here because it can help compare more signals, surface contradictions, simulate objections, consolidate qualitative evidence, and expose hidden assumptions before they become roadmap commitments.
But the value is not in the summary. The value is in the redesigned decision loop.
A better product operating model would connect customer signal, field feedback, support patterns, usage data, commercial priorities, strategic assumptions, and human product judgment into a more continuous evidence engine. It would make it harder for a loud request to masquerade as strategy. It would make it easier to ask what should stop being built. It would give product leaders more evidence before they make promises.
That requires human observation and accountability.
AI can process more signal than any product team could manually process. It can help expand the evidence base in ways that would have been unrealistic before. But it cannot sit in the customer conversation and notice the fear behind a polite answer. It cannot own the consequences of prioritizing one segment over another. It cannot carry the courage required to tell a senior stakeholder that their favorite idea is not supported by evidence.
This is where the Human Premium becomes more demanding, not less. I have written before about The Human Premium Enterprise, but Product makes the point concrete. When output becomes cheaper, the human premium moves toward judgment, observation, prioritization, and accountability.
AI does not give Product better judgment.
It gives Product the chance to stop wasting human judgment on work that machines can now help prepare.
Logistics Makes The Redesign More Disruptive
Logistics gives us an even more disruptive analogy because the value chain is so visible.
A simple AI use case in logistics might predict delays, optimize routes, forecast demand, automate warehouse reporting, flag exceptions, or help customer service explain shipment status. These are valuable improvements. They can reduce cost, improve responsiveness, and remove manual work.
But they are still band-aids if the logistics company remains built around delayed reports, separate planning cycles, fragmented handoffs, and reactive exception management.
The bigger question is what logistics becomes when the supply chain is treated less like a chain of reports and more like a living decision environment.
DHL’s Logistics Trend Radar highlights AI, advanced analytics, computer vision, and related technologies as major forces shaping logistics. That is useful, but the strategic implication is larger than adopting each trend separately.
The disruptive version is not better routing software.
The disruptive version is a logistics operating model where demand signals, port congestion, weather, supplier risk, warehouse capacity, transportation constraints, sustainability goals, customer promises, and financial exposure are continuously evaluated together.
- That changes the role of planning.
- It changes the role of exception management.
- It changes the role of procurement.
- It changes the role of customer communication.
- It changes who needs to be in the decision loop and when.
A company designed around delayed information has to create buffers everywhere. It adds inventory buffers, time buffers, approval buffers, meeting buffers, escalation buffers, and reporting buffers. Those buffers are not signs of stupidity. They are rational responses to uncertainty.
But if uncertainty can be reduced earlier and more continuously, then the operating model can change.
The old logistics model asks: how do we react faster when the chain breaks?
The AI-native question asks: how do we redesign the chain so fewer decisions depend on late recognition?
That is a much bigger move.
It is also more threatening. A route optimization tool does not challenge the organization very much. A living operating model that continuously rebalances decisions across functions does. It challenges planning cycles. It challenges the boundary between procurement and operations. It challenges local KPIs. It challenges who gets to decide when a tradeoff between cost, reliability, sustainability, and customer promise needs to be made.
That is why AI adoption becomes political.
Not because the tool is political, but because the redesigned value chain changes power, accountability, and decision rights.
Professional Services Should Feel Nervous
Professional services should not watch this from a safe distance.
The band-aid version of AI in consulting is already everywhere. Consultants can write decks faster. They can summarize interviews faster. They can draft proposals faster. They can generate research briefs faster. They can turn messy notes into structured findings faster. They can validate invoices faster. They can build first-pass operating models, transformation roadmaps, and stakeholder maps with less manual effort.
FINE!
But if that is all professional services does with AI, it is protecting the old economics with better tools.
The uncomfortable question is this: if consulting discovery can be compressed, then the old economics of selling large discovery phases become harder to defend.
That does not mean discovery disappears. It means discovery has to become more honest.
A lot of traditional consulting work sells confidence before evidence is strong enough. It sells a large statement of work because the client needs certainty, the consulting firm needs scope, and both sides participate in a ritual where the shape of the program becomes real before the value hypothesis has been properly tested.
AI changes that possibility space.
If cognition, synthesis, research preparation, qualitative comparison, and hypothesis generation become cheaper, then clients should expect more evidence earlier. They should expect sharper options before they commit. They should expect assumptions to be tested before transformation theater begins. They should expect professional services to reduce risk before asking for a large commitment.
That is not generic AI consulting. That is a different professional-services value chain, and one where the consulting partner has much more skin in the game.
It is also why I am increasingly willing to describe Open Source Consulting more explicitly as an experiment in AI-native professional services. The point is not to advise companies about AI from the outside of the old model. The point is to ask what professional services would look like if cognitive leverage were assumed from day one.
That means cheaper and faster value-hypothesis validation. More evidence before major transformation investment. Human judgment as the final validation gate. Less pressure to sell certainty too early. More transparency around whether an initiative is worth pursuing at all.
This is the open-heart-surgery version.
A generative agent that validates incoming invoices in procurement is a useful band-aid.
A professional-services model that validates added value before a statement of work is discussed is a redesign of the commercial relationship.
That is the difference.
And yes, that should make some professional-services people uncomfortable.
Not because AI replaces consultants, but because AI challenges which parts of consulting were valuable and which parts were protected by the cost of cognition. Your IP can be bought for a monthly subscription at OpenAI, Anthropic or Google, but your judgement, observations and instantaneous pattern recognition still provide unquestionable value to your clients.
The Old Operating Model Will Defend Itself
This is where leaders should be honest.
The old operating model will defend itself.
It will defend itself through budgeting. Through job titles. Through governance committees. Through procurement logic. Through KPIs. Through quarterly planning. Through transformation rituals. Through the idea that senior leadership still has the last word on what becomes redundant.
Periodic, cyclic goal setting by higher management is a good example.
It made sense in a world where information moved slowly, synthesis was expensive, and leadership needed scheduled moments to align the company. But AI has the potential to level parts of that playing field. More people can access more evidence. More teams can test assumptions. More functions can see patterns that previously had to be escalated, translated, filtered, and approved.
That does not remove the need for leadership.
It changes what leadership is for.
Leadership becomes less about being the final source of interpretation and more about designing the conditions under which better evidence can move through the organization without destroying accountability. It becomes more about objectives, constraints, decision rights, and judgment gates. It becomes more about knowing where human accountability must remain non-negotiable.
Microsoft’s 2026 Work Trend Index is useful here because it names the tension directly. Many AI users feel pressure to adapt, but many also feel safer focusing on current goals than redesigning work with AI. That is the transformation paradox in plain language: people can see the possibility, but the system still rewards the old behavior.
McKinsey’s 2025 State of AI research points in the same direction. The companies seeing stronger AI impact are not merely using more tools. They are much more likely to redesign workflows and define how human validation should work.
That is exactly the point.
The serious AI opportunity is not a better collection of disconnected use cases. It is a redesigned operating model where the company understands what AI prepares, what humans must observe, where judgment belongs, and who remains accountable.
Observation And Accountability Become More Important
The more AI expands cognition, the more valuable human observation becomes.
That may sound counterintuitive, but I think it is the heart of the matter.
AI can process patterns at a scale humans cannot. It can summarize more information, compare more alternatives, simulate more objections, and find more associations. It can help teams get to a better first question faster.
But it does not observe the world as a human does.
Observation is not just data capture. It is noticing what is strange. It is feeling the hesitation in a customer conversation. It is seeing when a stakeholder agrees too quickly. It is recognizing when the formal answer and the real answer are different. It is knowing that silence in a room means something. It is understanding context, fear, incentives, status, and trust.
AI also does not carry accountability.
Accountability requires consequence. It requires the possibility of being wrong in a way that matters. It requires respect for risk. It requires the courage to make a decision when the evidence is incomplete and the outcome has human, financial, or strategic cost.
AI lacks fear. That is part of why it can be useful. It can explore without embarrassment. It can generate alternatives without ego. It can challenge patterns without worrying about status.
But fear, in the right amount, is not only a weakness. It is part of respect. It is part of accountability. It is part of knowing that a decision matters.
The redesigned firm should not try to remove humans from accountability.
It should stop wasting accountable humans on work that no longer needs so much human limitation built into it.
That is the real Human Premium.
The Conscious Choice
AI is a challenge to the shape of the firm because it changes the assumptions beneath the firm.
- If expertise is easier to access, why are so many decisions still routed through old expert bottlenecks?
- If synthesis is cheaper, why are so many processes still organized around periodic reporting rituals?
- If evidence can be expanded earlier, why do so many transformation programs still sell certainty before validation?
- If field signal can become a living loop, why do product teams still receive it as filtered anecdotes?
- If logistics can become a continuous decision environment, why preserve planning systems built around delayed recognition?
- If professional services can validate value hypotheses before a statement of work, why defend the old discovery economics as if nothing changed?
These questions are uncomfortable because they do not attack the tool layer. They attack the inherited design of the company.
That is why the productivity conversation is too small. The question is not only how much faster the old company can move. The question is whether the old company is still the right shape.
If cognition is no longer scarce, the old value chain is no longer inevitable. It is a conscious choice.
But is it the right one?
