What is the real competitive advantage of AI in business?

The real competitive advantage in the AI era doesn't come from the AI technology itself. It comes from the transaction structure — how you design relationships, value flows, and incentives among all stakeholders. When competitors share the same business model, resources decide the winner. When models differ, the structure itself becomes the advantage.

Why did Duolingo succeed with AI while McDonald's struggled?

Duolingo embedded AI into its subscription-and-gamification model, creating new reasons for users to pay. McDonald's treated AI as a cost-saving tool bolted onto unchanged processes. The difference was not the AI model but the transaction structure around it. Duolingo redesigned stakeholder relationships; McDonald's did not.

What are the four competitive quadrants for AI in business?

The four quadrants are: (1) Efficiency Competition — same strategy and same model, where AI is a temporary efficiency multiplier; (2) Model Competition — same strategy but different model, where AI enables new transaction structures; (3) Platform Competition — different strategy but same model, where AI amplifies network effects; (4) Ecosystem Competition — different strategy and different model, where AI-native platforms create entirely new competitive landscapes.

How does transaction structure define the value of resources?

Resources do not have intrinsic value. The structure around them defines it. The same physical store is an asset in a traditional distribution setup but a liability in a direct-to-consumer subscription model. The same user data is a byproduct in a one-off transaction and a goldmine on a multi-sided platform. Value, rarity, and imitability are properties of the structure, not the resource itself.

What are the four implementation paths for AI-native competition?

The four paths are: Enablement (using AI as a tool while restructuring external relationships), Form First (building end-to-end teams organized around value delivery), Spirit First (letting AI become the independent agent in key processes), and Integration (human-AI-agent tri-symbiosis with self-sustaining value, capability, and rule-evolution loops). Each stage can build a defensible moat.

What mindset shifts do leaders need for AI-driven business model innovation?

Leaders need three shifts: (1) Stop the tech arms race — models and talent are purchasable, what matters is embedding AI into a better transaction structure; (2) Start model innovation at any stage — don't wait for full organizational transformation; (3) Shift from endowment thinking to design thinking — ask what transaction structure creates the most value, then find the resources to support it.

Can companies build a moat without reaching full AI-native integration?

Yes. Companies do not need to reach the highest stage of AI integration to build a defensible moat. Even at the Enablement stage, where AI is used simply as a tool, the right transaction structure can create a deep enough competitive barrier. What matters is structural differentiation, not technological sophistication.

What is the single most important question every company should ask about AI?

The most important question is: If every competitor has the same AI, what keeps us ahead? The answer lies not in technology but in how you redesign stakeholder relationships, value flows, and incentive rules within your business model. The most successful companies of the next decade will not be the ones with the most AI resources, but the ones that used AI to rebuild their structure first.

How is AI changing business model competition?

AI lowers the cost of designing and deploying new business models, shifting competition from resource-based advantages to structure-based advantages. When business models are similar, AI acts as a resource amplifier, and traditional advantages like brand and supply chain matter. When business models differ, AI becomes a structural reinventing force, and the transaction structure itself becomes the barrier to imitation.

When Everyone Has Access to the Same AI, What Actually Builds a Moat?

Q: What is the difference between level one and level two capabilities in the AI era?

Level one capability is the ability to deliver value within a given structure — supply chain, brand, execution. Its value is defined by the structure it sits in. Level two capability is the ability to design and reconstruct the transaction structure itself — seeing where the old model tops out, imagining a new one, and iterating. Level two is what Teece called 'dynamic capabilities' and is the rarest skill in the AI era.

The same AI tools. Completely different outcomes. Why?

Duolingo plugged GPT-4 into its platform in 2023 and launched Duolingo Max. Nothing groundbreaking there — plenty of companies had access to the same model. But instead of using it as a chatbot bolted onto the side, they wove it into their subscription-and-gamification model. AI-powered roleplays. Smart answer explanations. Suddenly the AI became a reason to pay, not just a feature to play with. User engagement went up. Paid conversion went up.

McDonald’s went the other direction. They spent heavily on AI — menu recommendations, voice ordering, store systems. But most of it was slapped onto existing processes. The way McDonald’s interacts with customers, franchisees, and suppliers barely changed. The tech was there. The results weren’t.

The gap isn’t about who has better AI. It’s about who has figured out how to structure around it.

The Real Divide Isn’t Technology. It’s Structure.

Most companies treat AI as a way to make existing operations a bit faster and cheaper. That’s not wrong — it’s just not nearly enough. When every competitor can tap the same foundation models (especially after DeepSeek open-sourced top-tier models in early 2025), the technology stops being a differentiator. What actually matters is what you build around it.

Here’s the thing that keeps gnawing at me: sustainable advantage in the AI era doesn’t come from AI itself. It comes from the transaction structure — how you design relationships, value flows, and incentives among everyone involved. When two companies compete with the same model, resources decide the winner. When their models differ, the structure itself becomes the moat.

Duolingo won because its transaction structure was different: subscribers pay for AI-enhanced learning experiences, not for raw model access. McDonald’s struggled because the structure stayed the same — AI was just a cost-saving patch on an unchanged system.

Four Ways AI Changes the Game

Whether AI gives you a real edge depends on where you sit on two axes: competitive strategy (same or different) and business model (same or different). That gives us four quadrants.

1. Efficiency Competition — Same Strategy, Same Model

Think two supermarkets on the same street. Same model, same playbook. AI is a straightforward efficiency multiplier — optimize scheduling, reduce waste, predict demand. The problem? Your competitor can do the same thing with the same tools. Short-term cost advantages get competed away. The winner is whoever had more resources from the start — better supply chains, stronger brand, deeper pockets.

The classic Resource-Based View holds true here. AI doesn’t create a lasting gap. It just widens and then narrows back.

2. Model Competition — Same Strategy, Different Model

Two companies serve the same small business tax customers. One bills by the hour. The other uses AI for standardized compliance checks and charges a yearly subscription. Same customers, totally different transaction structures.

Here, AI isn’t a patch — it’s the reason the new model exists. And once that model runs, copying the AI is useless. You can’t replicate the pricing logic, the customer segmentation, or the profit-sharing rules just by buying the same API key. This is exactly how resource-poor newcomers overtake incumbents.

3. Platform Competition — Different Strategy, Same Model

Food delivery and local deals are different businesses, but they share the same underlying model: a platform connecting supply and demand, taking a cut. AI amplifies the scale advantage — better matching means better experiences, which pull in more merchants, which generate more data, which makes AI even better at matching. A loop that feeds itself.

The moat isn’t the AI. It’s the whole coupled system.

4. Ecosystem Competition — Different Strategy, Different Model

AI-native platforms — think research collaboration tools or personalized education systems — don’t just compete differently. They compete in a different game altogether. First movers define the rules, set user habits, and build data flywheels that didn’t exist in the old framework. These are moats the old models couldn’t even see.

The pattern here: the more similar the business model, the more AI acts as a resource amplifier and traditional resource advantages matter. The more different the model, the more AI becomes a structural reinventor. Homogeneous models produce gradient advantages — you can always catch up by spending more. Only unique transaction structures produce real, defensible moats.

Why Structure Defines Value

This is where I think most strategists get it wrong. Resources don’t have intrinsic value. The structure around them does.

Put the same physical store in two different models: in a traditional distribution setup, it’s an asset. In a direct-to-consumer subscription model, it’s a liability. The same user data in two different structures: in a one-off transaction, it’s a byproduct. On a multi-sided platform, it’s the whole reason you’re in business.

This flips the conventional logic. “Valuable, rare, hard to imitate” — these aren’t properties of the resource itself. They’re properties of the structure the resource lives in. When the dominant transaction structure shifts, yesterday’s core asset becomes today’s dead weight. Not because the asset changed. Because the rules around it did.

Real moats are structural. You can buy the same AI. You can hire the same people. You can benchmark the product. But you cannot copy an entire transaction structure overnight — because it’s not a single point. It’s a web. Changing it means renegotiating every stakeholder relationship, every rule, every incentive. The three classic barriers to imitation — causal ambiguity, social complexity, path dependence — these aren’t features of isolated resources. They are system-level properties of multi-stakeholder structures.

That’s why business model innovation runs deeper than product or technology innovation. Get the structure right, and ordinary technology builds a moat. Get it wrong, and the most advanced AI becomes just another cost line on the P&L.

Four Paths to AI-Native Competition

The relationship between humans and AI evolves through three stages: tool, partner, and integrated agent. You don’t need to reach the final stage to build a moat. Each stage offers its own structural opportunities.

Path 1: Enablement — Tool First, Structure First

You don’t need to change your internal architecture to change your external transaction structure. One major retail chain uses the same LLMs as everyone else for assortment planning and scheduling. But they repurposed the user data AI helped analyze — instead of buying and selling for margin, they now provide scenario-based solutions and take a cut of the results. Competitors can buy the same AI. They cannot replicate the profit-sharing mechanism and the scenario-specific data that comes with it. That data isn’t special because of its volume. It’s special because of who it comes from and how it gets used inside this particular structure.

Path 2: Form First — Build the End-to-End Skeleton

Tear down departmental walls. Organize teams around value delivery, end to end. AI stays a tool, called when needed. The custom furniture industry is a good example — traditionally fragmented across design, production, and installation. Some companies now run full-cycle teams responsible from measurement to follow-up. AI is available to everyone. But the team’s authority structure, coordination rules, and incentive design? Those are close to impossible to copy. Delivery time and customer satisfaction advantages don’t vanish when competitors buy the same model.

Path 3: Spirit First — AI Becomes the Agent, Structure Follows

AI takes over key decision-making roles and owns outcomes. The organization still runs on traditional divisions. A common example: e-commerce refund processing. The AI handles it end to end — real-time judgment, instant payout, later settlement. This fundamentally restructures the tri-party relationship among platform, consumer, and merchant. Other platforms can build similar systems. They cannot replicate the three-party payout rules and the risk data flywheel that gets smarter with every transaction. Every payout teaches the system something new. That’s a loop that compounds.

Path 4: Integration — Tri-Symbiosis

At the highest level, the organization is fully end-to-end. Humans, AI agents, and intelligent entities coexist in a dynamic system. Value delivery, capability growth, and rule evolution run on three self-sustaining loops. Coordination happens through shared internalized rules rather than top-down commands or case-by-case negotiations. Hierarchy dissolves into on-demand collaboration.

Important caveat: not every company needs to reach this stage. Even at the enablement stage, the right transaction structure builds a deep enough moat. Don’t let perfect be the enemy of structural.

Three Mindset Shifts Every Leader Needs

First, stop the tech arms race. Models, compute, and AI talent are all purchasable and replicable. They don’t build moats. What matters is who can embed AI into a better transaction structure. If your competitive strategy starts with “which model should we use,” you’re already losing.

Second, start model innovation at any stage. Don’t wait until your organization is fully transformed or your technology stack is mature. The earlier you look for structural differentiation, the sooner you escape the commodity trap. Waiting is the most expensive strategy.

Third, shift from endowment thinking to design thinking. Stop asking “What resources do we have?” Start asking “What transaction structure would create the most value?” Then figure out what resources you need to make it work — and go find them, borrow them, or partner to get them. Resources follow design. Not the other way around.

One more thing about capabilities — because this gets confusing. There are two levels. Level one is the ability to deliver value within a given structure — supply chain, brand, distribution, execution. Its value is defined by the structure it sits in. Level two is the ability to design and reconstruct the transaction structure itself — seeing where the old model tops out, imagining a new one, and running the iteration loop. This is what Teece called “dynamic capabilities.” It’s the engine that generates better structures. And in the AI era, it might be the rarest thing in the room.

The Question That Matters

For most organizations today, the real question isn’t “Should we use AI?”

It’s this: if every competitor has the same AI, what keeps us ahead?

I don’t think the answer will come from the engineering team. It’ll come from how you redesign the stakeholder relationships, the value flows, and the incentive rules that make up your business model. The most successful companies of the next decade won’t be the ones with the most AI resources. They’ll be the ones that used AI to rebuild the structure first — because technology eventually becomes table stakes, but structure determines who gets the value.

The moat isn’t in the model. It’s in the architecture around it.