This is my second article that draws out the implications of the 2026 Stanford AI report. This article is an update. The first article can be found here
Most executives are still asking the wrong AI question.
They ask: How fast should we adopt? Which use cases matter? Where do we deploy agents? How do we automate work?
Those are fair questions. They are no longer the main ones.
The Stanford AI Index 2026 suggests that AI is already spreading fast enough, delivering enough practical value, and improving quickly enough on task performance that the larger issue is no longer whether firms adopt AI.
The larger issue is what logic AI is now scaling. Generative AI reached 53% population adoption within three years, estimated annual consumer value in the U.S. reached $172 billion by early 2026, organizations reported 88% AI adoption, and Stanford says agent systems are moving from answering questions toward completing tasks.
That is the strategic shift.
AI is no longer just a side tool for productivity. It is becoming part of the operating fabric of markets. But if that is true, then the real danger is not slow adoption.
It is fast adoption of the wrong strategic model.
AI does not create market logic from nothing. It learns from and scales the assumptions firms already embed in how they define the category, map demand, position brands, structure portfolios, orchestrate interactions, and measure success. If those assumptions are weak, AI does not fix them. It amplifies them.
That is why I am uneasy with the phrase AI strategy. Most companies do not need a separate AI strategy deck full of pilots, tools, and agents disconnected from the business. They need to understand how AI changes the system they are in, and then ask a harder question:
What strategic logic are we now encoding into machines?
There is another reason this matters.
Much of the current conversation about AI and brands is focused on visibility: GEO, answer-engine optimization, machine-readable content, structured product data, discoverability in AI search. That is understandable. There is important work done in this area by a number of thought leaders. Thomas Marzano for example speaks of the legible-loveable standard.[1] Jochen Sengpiehl calls brand visibility the key discipline for the 21st century.[2] BCG’s work on discoverability argues that brands increasingly need to be selected, summarized, and surfaced by AI systems, and OpenAI’s shopping products are explicitly designed to help users discover, compare, and decide on products conversationally.[3]
All of that matters. But it does not go deep enough.
Visibility is not the deepest strategic issue. Visibility is the downstream expression of a deeper system. Before a machine can surface a brand, it is already interpreting the category, structuring demand, deciding which criteria matter, framing the choice set, and ranking what deserves to be seen. That is why discoverability is necessary but insufficient. The more important question is not only whether the machine can find the brand, but what market logic the machine is using to decide what deserves to be found. This is the real move from visibility to selection logic.
That is where the Stanford report becomes useful. It does not speak in the language of category meaning, demand spaces, or brand architecture. But it does describe the conditions under which those questions become unavoidable. AI is reaching market scale. Consumers are deriving real value from it. Organizations are integrating it broadly. Agents are improving in task execution. At the same time, Stanford stresses that our ability to measure and manage AI is lagging behind capability.
So the management problem is not simply deployment. It is whether the firm is giving AI the right strategic architecture to run.
AI will scale the whole system, not just the output.
The right way to think about AI is not as a content engine, and not even just as a discovery engine. It is increasingly a system that can scale strategic logic across five connected layers:
- Category meaning
- Demand structure or demand logic
- Positioning plus portfolio and brand architecture
- The interaction layerCompound economic profitIf those layers are coherent, AI can amplify a strong system. If they are weak, AI can industrialize confusion. You probably read about these five steps before. They are a model of brand performance that I have articulated earlier, see here.
I believe we need to move toward this five-point brand performance model so that we peg AI at the right level, the strategy level, or else we just optimize execution.
That is the deeper risk.
If the category is defined too narrowly, AI will narrow demand the wrong way.
If the demand structure is misunderstood, AI will optimize the wrong problem faster.
If positioning is vague, AI will make the brand easier to compare but not easier to choose.
If the portfolio is confused, machine-mediated recommendation will expose that confusion sooner.
If the interaction layer is weak, AI will accelerate friction rather than remove it.
If the economics are shallow, AI may improve efficiency while weakening pricing power, returns on capital, and long-term advantage.
That is why the real issue is no longer adoption. It is strategy. It is logic.
1. Category meaning
The first layer is category meaning.
What is the category fundamentally about? What kind of problem is it solving in people’s lives? What game is being played?
This matters because AI cannot guide a market well unless it has some working model of what the category is for. If that model is narrow or wrong, the system will organize the choice set the wrong way.
Airbnb is useful here because it shows what a category move looks like. The company has expanded beyond a narrower lodging frame toward a broader travel-and-stay system, adding Services and redesigning Experiences into the same app. That is not just product expansion. It is a change in category meaning. Airbnb is not only helping people book a place to sleep. It is trying to organize more of the trip around the stay.
And this is not happening inside a small or speculative business. In fourth quarter 2025, Airbnb reported revenue up 12% year over year, gross booking value up 16%, and nights and seats booked up 10%. That matters because it shows that broader category meaning is not just intellectual positioning. It can sit inside a scaled and growing business system.
That is the broader point. Before a machine can recommend, compare, or guide action, it has to inherit some category frame.
So the first question is not just, can AI find the brand?
It is: what category game is AI helping the customer play?
2. Demand structure or demand logic
But category meaning is still too broad on its own.
The second layer is the structure of demand inside that category.
Not demand as volume. Demand as context, moment, need, trade-off, friction, and occasion.
Instacart is one of the clearest examples of this shift. Ask Instacart did not begin with a shelf, a keyword, or a brand. It began with a consumer problem: what is for dinner, what fits dietary needs, what works on a budget, and what ingredients are needed. The entry point is no longer only the product category. It is the demand moment itself.
That is demand logic in action.
The consumer is not entering through “sauce,” “snacks,” or “brand X.” The consumer is entering through dinner, convenience, cost, dietary restriction, or household need. The system then organizes supply around that moment.
And again, this is not a niche example. Instacart reported first-quarter 2026 GTV of $10.29 billion, up 13% year over year, revenue of $1.019 billion, up 14%, and adjusted EBITDA of $300 million, up 23%. That does not prove demand logic alone caused the result. But it does show that better interpretation of demand and better orchestration of supply have real economic weight at scale.
This is why demand spaces matter more, not less, in an AI-mediated market. If the demand map is wrong, AI will simply optimize the wrong map faster.
3. Positioning plus portfolio and brand architecture
This is the layer that often gets reduced too quickly to positioning alone. It should not.
Positioning matters, but this layer is really where category meaning and demand logic are translated into a structured supply system. That means brand positioning, portfolio design, and brand architecture all belong here.
This is where the firm decides:
What role should this brand play?
Which demand spaces should it serve?
How should it differ from sibling brands?
How should the portfolio map across the demand structure?
What supply system best fits the demand system?
This layer is critical because it does not merely reflect demand logic. Over time, it can stabilize, sharpen, or reshape category meaning itself.
That is the feedback loop many firms miss.
Category meaning defines the broad game.
Demand logic structures that game into real spaces.
Portfolio and architecture map supply into those spaces.
And once that supply system is established, it starts to teach consumers, channels, experts, and increasingly machines what the category is actually for.
That is why portfolio and architecture are not downstream housekeeping. They are one of the mechanisms through which category meaning itself gets reinforced.
La Roche-Posay is a very good example. I would not say La Roche-Posay completely changes category meaning in the way Airbnb broadens travel. But it clearly helps sharpen and legitimize a particular category frame: dermatological beauty, science-backed skin health, and medically credible care for real conditions. L’Oréal explicitly places it inside Dermatological Beauty and describes the division through healthcare partnerships, medical credibility, and prescribed solutions. The division also reported 25.8% market share in 2025 and said it reached more than 320,000 healthcare professionals.
At the same time, La Roche-Posay is built around a demand structure that clearly favors its strengths: acne, hyperpigmentation, barrier repair, sun care, and sensitive skin. L’Oréal’s reporting on Mela B3, based on a global epidemiological study of pigmentary disorders covering 48,000 people across 34 countries, is a strong example of demand-structure work, not just product marketing.
So the strategic point is this:
La Roche-Posay does not simply fit an existing category and demand map. It helps codify a more medical, condition-based demand structure, and through its role within L’Oréal’s Dermatological Beauty portfolio, it reinforces a sharper meaning for the category itself. That is Layer 3 doing its real work.
4. The interaction layer
This is the layer most often misunderstood. Strategy does not move directly from portfolio to economics. It moves through the market-facing operating layer: the full interaction system through which the firm meets demand.
This layer is broader than “execution” in the narrow sense, as I defined it before. It includes classic brand building through impressions. It includes search and retail visibility. It includes product pages, service flows, recommendations, diagnostics, and routines. It includes channels, touchpoints, customer experience, and agent-mediated assistance. And increasingly, it includes SRAL loops: Sense, Reason, Act, Learn.
That is why Layer 4 should be called the interaction layer. I have written about this before as well.
Traditional impression-based brand building absolutely belongs here. In stable mass markets, the interaction layer may lean heavily on reach, frequency, salience, and physical and mental availability. In more dynamic or AI-mediated markets, it expands to include sensing intent, reasoning across context, recommendation, assistance, and learning.
This is exactly where AI matters most concretely.
AI is entering the interaction layer itself. It helps sense signals, reason across options, act through recommendations or next best actions, and learn from behavior and feedback.
Instacart is helpful here as well. Its value is not only that it identifies a demand moment. It also orchestrates the interaction from need recognition to product selection to basket creation. That is a richer form of go to market than media or shelf presence alone. And because this orchestration now sits inside a platform with over $10 billion in quarterly GTV, it is economically meaningful, not just conceptually elegant.
La Roche-Posay is again useful here. SPOTSCAN+ is not just an AI gimmick. It requires three selfies, uses a bank of 6,000 scientific images across skin types, and relies on a dermatologist-recognized acne severity scale to analyze skin and guide recommendations. It links a consumer problem, an assessment layer, a recommendation pathway, and a treatment routine. It sits inside the interaction system, not just the communications system. That is why it is more interesting than generic ad automation. It shows AI strengthening the bridge between problem recognition and action.
This is also why much current “Brand OS” thinking still feels too narrow. It is often about prompt governance, brand consistency, templates, and content control. Useful, yes. But if it lives only at the expression layer, it misses the bigger issue. The real question is whether the business has a stronger operating system for category meaning, demand logic, positioning, architecture, and interaction.
If those upstream layers are wrong, a more efficient Brand OS will only produce more coherent irrelevance.
5. Compound economic profit
The last layer should not be described loosely as “performance.” It is much more specific than that. The final layer is whether the whole system produces economic profit and whether that economic profit can be compounded over time.
This is where the model has to go all the way to ROIC, EVA, or economic profit, not stop at growth, share, or campaign metrics. This matters to me because there is a permanent discussion about broadening the role of marketing today. I agree this is necessary, however, it also is only useful if marketing assumes more accountability for value creation and overall business results. See here.
A firm can grow and still destroy value.
A firm can improve efficiency and still weaken pricing power.
A firm can automate execution and still reduce returns on capital.
That is why the end point has to be sharper.
The question is not simply: Did the system perform? It is: Did the system create returns above the cost of capital, and can those returns be reinvested to strengthen future advantage?
That is the compound economic profit loop.
Better category meaning leads to better demand fit, which leads to stronger positioning and portfolio logic, which leads to better interactions, which leads to stronger willingness to pay, mix, margin, and returns, which generates economic profit, which can then be reinvested in better science, better experiences, stronger brands, and future demand spaces. That is the compounding logic.
La Roche-Posay works as an example because the economic layer is not separate from the first four. L’Oréal’s reporting says La Roche-Posay remained the number one growth contributor in Dermatological Beauty in 2025 and significantly outpaced the market. That is not brand-level ROIC disclosure, but it is strong directional evidence that the integrated system is economically powerful.
The strategic point is not that La Roche-Posay is good at AI. It is that La Roche-Posay has built a coherent system across meaning, demand, role, interaction, and economics. AI then becomes an amplifier of that system, not a substitute for it.
What the examples prove
These examples matter because they connect the five layers to visible business outcomes.
Airbnb shows that a broader category frame can support portfolio expansion and growth. In Q4 2025, that showed up in 12% revenue growth and 16% gross booking value growth.
Instacart shows that demand logic can become the entry point to a very large transaction system. In Q1 2026, that system produced $10.29 billion in GTV, $1.019 billion in revenue, and $300 million in adjusted EBITDA.
La Roche-Posay shows that when category meaning, demand structure, portfolio role, interaction design, and economics reinforce one another, the result is not just better execution but a stronger business system. L’Oréal’s own reporting that it was the number one growth contributor in Dermatological Beauty is exactly the kind of evidence that matters here.
That is why this is not a purely conceptual model. It is a model of brand and business performance.
What Stanford adds
The Stanford AI Index 2026 does not prove this framework. That is not its role.
What it does is make the timing unmistakable.
AI is diffusing at historic speed. Consumers are getting real value from it. Organizations are integrating it broadly. Agents are moving from answers toward tasks. And governance, transparency, and institutional readiness are lagging. That combination is what makes this moment strategic.
The problem is no longer just whether a firm uses AI. The problem is whether it is giving AI the right system to scale.
That is why the first management question should no longer be: How do we deploy AI?
It should be: What market logic are we asking AI to run?
What category meaning are we teaching the system?
What demand map are we asking it to optimize?
What role should each brand play in that map?
What architecture should organize the supply system?
What interaction logic are we orchestrating through impressions, experiences, recommendations, and SRAL loops?
What economic profit should follow if the system works properly?
These are harder questions than use cases. But they are now the real strategic questions.
The real risk
The real danger is not failing to adopt AI. It is adopting AI before you are clear about the strategic model you want it to scale. If the strategic logic is strong, AI can help firms sense better, reason better, act faster, and learn more effectively. If the strategic logic is weak, AI will industrialize the weakness.
That is why the firms that win will not simply be the ones with the most agents, the most tools, or the fastest pilots. They will be the ones that understand a simpler truth:
AI will increasingly scale whatever strategy you give it.
And that is why strategy matters more, not less, in the age of AI.
[3] https://www.bcg.com/x/the-multiplier/the-future-of-discoverability?utm_source=chatgpt.com