Your MBA is lying to you
Your MBA is lying to you
Porter's Five Forces, the BCG Matrix, SWOT, and the rest of the strategy canon don't just fail in the AI era. They actively produce the wrong answers. And the people most fluent in them are the most confidently wrong.
Somewhere this week, a strategy team at a Fortune 500 company will open a slide deck, populate a Porter's Five Forces diagram, and use it to make a major investment decision. They will do this rigorously. They will do it with conviction. And in the AI era, they will very likely arrive at the wrong answer.
This isn't the usual complaint that business school frameworks are "a bit outdated" or "need refreshing." The argument here is sharper and more uncomfortable: the canonical strategy frameworks taught in every MBA program on earth have become actively dangerous. Not incomplete. Not imperfect. Dangerous, in the specific sense that they give intelligent people a structured, sophisticated-sounding basis for making choices that are fundamentally wrong for the competitive dynamics of artificial intelligence.
To understand why, we need to see what these frameworks actually do to the people who use them, and then build what should replace them from scratch.
Part IFrameworks aren't tools. They're cognitive operating systems.
Here is the generous reading of strategy frameworks: they're analytical tools. You pick one up, apply it, put it down, and combine its output with other inputs. That's how they're described in textbooks.
Here is the accurate reading: they're mental models that, once internalized, determine what you can see. An executive who has spent two years at a top business school and fifteen years in strategy consulting doesn't "use" Porter's Five Forces the way a carpenter uses a hammer. They think in it. It becomes the unconscious structure through which they interpret competitive reality. Industries have boundaries. Rivals are identifiable. Threats come from five directions. Competitive advantage comes from positioning within that structure.
This is fine when the model matches reality. It's catastrophic when it doesn't.
Think of it this way. Imagine you're using a map of London from 1990 to navigate London in 2025. Most of the streets are still there. Many landmarks are the same. The map feels useful. But the Jubilee line extension doesn't exist on it. The entire Docklands redevelopment is blank. If you're trying to get to Canary Wharf, the map isn't just unhelpful; it's telling you there's nothing there. A person with no map might ask for directions. A person with the old map is confident they already know the territory.
That's the situation with legacy strategy frameworks in the AI era. The territory has changed. The maps haven't. And the people carrying the maps are the most senior, most credentialed, most confident people in the room.
Part IIFive autopsies
Let's be specific. Here are the five most widely taught strategy frameworks, each examined not for where they fall short, but for the specific wrong answer they produce when applied to AI-era competition.
What it assumes: Industries have boundaries. Competition happens within those boundaries along five vectors: rivalry, new entrants, substitutes, supplier power, buyer power.
The wrong answer it produces: Consider a regional insurance company running a Five Forces analysis. They'd examine rival insurers, the threat of new insurance startups, substitute products (self-insurance, captives), broker bargaining power, and customer price sensitivity. This is a thorough analysis. It would probably recommend investments in underwriting efficiency and broker relationships.
What it would never surface: the possibility that an AI orchestration layer, a Claude or GPT-based agent that shops, compares, negotiates, and switches policies autonomously on behalf of the customer, could make the insurer's brand invisible. The agent doesn't care about your brand. It cares about price and coverage fit. The existential threat isn't from another insurer. It's from a layer that doesn't appear on Porter's map, because Porter's map assumes the industry boundary is the unit of analysis.
What it assumes: Portfolio allocation should be based on market growth rate and relative market share. Milk cash cows, invest in stars, fix or kill question marks, divest dogs.
The wrong answer it produces: A media company's content licensing division is a classic cash cow: high share, low growth, throwing off cash. The BCG playbook says harvest it. But what if generative AI is flooding the content market, collapsing the scarcity value that licensing revenue depends on? That "cash cow" isn't stable. It's a melting ice cube. BCG's matrix sees current cash flow and calls it durable. It has no mechanism for asking: "Is the reason this asset generates cash about to evaporate?"
Meanwhile, the company's early AI initiative gets classified as a "question mark" and receives the patient, wait-and-see posture the matrix recommends for that quadrant. But AI ventures don't reward patience. Data compounding means the first company to get the flywheel spinning pulls away permanently. The BCG matrix's tempo is wrong. It permits a "let's see how this develops" approach to a dynamic where hesitation is permanent lockout.
What it assumes: That the competitive environment is stable enough to inventory. You list your strengths and they remain strengths. You identify threats and they arrive from identifiable directions.
The wrong answer it produces: A customer service company lists "5,000 trained agents across three continents" as a strength. On a Tuesday, it is one. By Thursday, a competitor has deployed AI agents that handle 80% of the same inquiries at a tenth of the cost. The "strength" has become a structural liability: expensive, slow to retrain, and contractually locked in.
SWOT doesn't distinguish between strengths that compound and strengths that erode. A strong brand compounds. A large workforce of manual processors erodes. SWOT puts them on the same list, in the same column, with the same implied stability. An executive staring at a SWOT grid has no mechanism for asking: "Which of these strengths are actually ticking time bombs?"
What it assumes: A business is a chain of primary and support activities. Competitive advantage comes from performing individual links better or more cheaply than competitors.
The wrong answer it produces: A law firm analyzes its value chain: intake, research, drafting, review, client delivery. It invests in making the research link faster and the drafting link cheaper. Sensible, by value chain logic. But AI doesn't optimize links. It collapses them. AI-assisted legal research isn't "better research." It's the elimination of research as a distinct, billable activity. The value chain framework assumes the chain's structure is fixed and the question is "how do we perform each link better?" The AI-era question is: "Which links survive at all?"
An executive running value chain analysis will hire consultants to optimize a process that a model will handle end-to-end within eighteen months. The framework pointed their investment at work that is about to disappear.
What it assumes: That advantages, once built, can be defended for extended periods through barriers to entry, switching costs, or proprietary resources.
The wrong answer it produces: A SaaS company builds a sophisticated data analytics dashboard over three years. It has proprietary visualizations, deep integrations, and a loyal customer base. Classic sustainable advantage. Then a frontier model enables any user to ask questions of their data in plain English, rendering the carefully built interface a convenience rather than a necessity. The "sustainable" advantage had an expiration date that nothing in the traditional analysis could have detected, because the threat wasn't a competitor building a better dashboard. It was a general intelligence making dashboards less necessary.
The very concept of "sustainable" is misleading in a world where capability overspill from general AI systems can flood your niche overnight, without any competitor deliberately targeting you.
The pattern across all five is the same. Each framework takes something that's genuinely unstable, an industry boundary, a cash flow stream, a workforce, a value chain link, a competitive moat, and treats it as fixed. Each produces a recommendation based on that false stability. And each gives the executive a structured, defensible rationale for the wrong investment.
A new map: The AI power matrix
If legacy frameworks are wrong because they analyze horizontal competition within bounded industries, the replacement needs to start from a different geometry entirely. The AI era's defining power struggles happen vertically, between layers of a technology stack, and the winners are whoever controls the scarcest bottleneck at any given moment.
Two axes matter. Scarcity: could someone else provide what you provide? And ecosystem gravity: has the industry organized itself around you in ways that are painful to undo? These produce four archetypes.
Quarries are the strongest position: you own something scarce AND the ecosystem is built on top of you. NVIDIA is the purest example; not just scarce chips, but CUDA as a software lock-in that makes switching feel like rewriting a decade of work. Wellsprings have the scarce capability (frontier AI models) but haven't locked the ecosystem in yet. Deltas control the convergence point; what they offer isn't unique, but all traffic flows through them. And Driftwood is the honest quadrant; nothing scarce, no gravitational pull. If your model provider ships your feature natively, what do you have left?
The matrix matters because it asks two questions that legacy frameworks never do: What layer do you occupy? and Is your position in that layer durable?
Part IVSix currents: replacing Porter for the AI era
Porter gave us five forces that act horizontally, within a bounded industry. The AI era needs a framework for forces that act vertically, diagonally, and sometimes in directions that don't map to traditional competitive analysis at all. I call these currents rather than forces, because they're directional, dynamic, and constantly shifting. You don't position against a current. You navigate it.
Layer collapse
Can the layer above or below you absorb what you do? This is the defining competitive dynamic of the AI era, and Porter has no equivalent. It's not "threat of new entrants," because nobody is entering your market. The layer itself is being swallowed.
When OpenAI adds a native code interpreter, every startup that built a code-execution wrapper doesn't lose market share. It ceases to exist as a category. When a model provider adds built-in web search, a dozen search-augmentation companies evaporate. No competitor killed them. The layer collapsed.
The commodity tide
Every month, the floor of "what a cheap, widely available model can do" rises. This creates relentless erosion of any capability that isn't at the absolute frontier. What was a breakthrough last January is a commodity by September.
Think of it as a rising tide in a harbor. Your boat (your product's differentiator) is anchored at a fixed height. As the water rises, your boat doesn't get higher; it just has less distance between it and everything else floating nearby. DeepSeek proved that even cost barriers don't hold: frontier-competitive capability can emerge at a fraction of expected cost. Open-weight releases like Meta's Llama are deliberate accelerants of this tide, used by companies that benefit from commoditizing the model layer.
Substrate stacking
Every layer you depend on can independently extract rent from you. A platform-era SaaS company had one real dependency: its cloud provider. An AI-native company depends on chip manufacturers, cloud providers, model providers, and possibly data licensors, all stacked vertically. Each can change pricing, rate-limit you, alter terms, or deprecate capabilities unilaterally.
It's like renting an apartment where the landlord, the power company, the water utility, and the internet provider can each independently double their rates or cut off service. One disruption at the bottom of the stack propagates through every layer. Porter treats supplier power as a single variable. In the AI era, it's a compounding risk function.
Data compounding
Does usage make you harder to displace? Traditional network effects were about connections between users (more people on a phone network makes the network more valuable). AI network effects are about a quality improvement loop: more usage generates more behavioral data, which improves the model, which attracts more usage.
This is a flywheel, and once it's spinning, it's brutally hard for competitors to match. The advantage isn't in your code or your team. It's in accumulated signal from millions of interactions. This explains why distribution matters so much: getting to users first isn't about revenue. It's about getting the flywheel spinning before anyone else.
Capability overspill
The risk that a general system suddenly becomes competent in your specific domain, without anyone planning for it. GPT-4 didn't set out to replace legal research tools. It just turned out to be good at legal reasoning. No one at a foundation model lab decided to target financial analysis. The model simply became competent at it.
This is fundamentally different from anything in Porter. There's no competitor to track, no market entry to monitor. The "threat" is an emergent property of a system built for entirely different purposes. It's like owning a specialty shop in a town where a river is gradually widening; the river isn't targeting your shop, but one day the water reaches your doorstep.
Orchestration capture
Who controls the surface the user actually touches? The orchestration layer decides which model handles which query, which provider gets the traffic, which capability the user even knows exists. It's a chokepoint that extracts value from every layer beneath it.
Apple doesn't need a frontier model. It needs 2 billion devices where the user's first interaction happens. Microsoft doesn't need the best AI. It needs to embed AI into Office and Windows, where the work already lives. Porter's "buyer power" assumed buyers chose between competing products. Orchestration capture is different: the orchestrator pre-selects on the buyer's behalf, often invisibly. The user doesn't choose between Claude and GPT. The orchestrator routes the query before the user knows there's a choice.
The geometry is different
What makes these six currents fundamentally different from Porter isn't just updated examples. It's a different geometry. Porter's five forces all act on the same plane: horizontal competition within a bounded industry. The six currents act vertically, diagonally, and sometimes in directions that have no traditional analog. Layer collapse is vertical. Capability overspill is orthogonal. The commodity tide is environmental. Substrate stacking is structural.
They don't converge on Porter's question: "How profitable is this industry?" They converge on a different question entirely: At which layer does value accumulate, and how durable is that accumulation?
What should replace framework-driven strategy?
The answer is not a better framework in the same genre. It's a different kind of strategic thinking entirely: one that starts from layers rather than industries, treats competitive boundaries as fluid rather than fixed, and asks "where does value accumulate in a world where capability is abundant and general?" rather than "how do I position within my industry?"
Three shifts are required.
1. From industry analysis to layer analysis
Stop asking "what industry am I in?" and start asking "what layer of the stack do I occupy?" Then ask the harder question: is that layer thickening (becoming more valuable and defensible) or thinning (becoming commodity infrastructure for the layer above)?
2. From positioning to navigation
Legacy strategy is about finding a position and defending it. AI-era strategy is about sensing which direction the currents are flowing this quarter and maintaining the organizational agility to reposition continuously. This isn't a comfortable metaphor for executives who want a three-year plan. But a three-year plan built on a three-month reality is worse than no plan at all.
3. From competitive intelligence to layer intelligence
Traditional competitive intelligence asks: what are my rivals doing? Layer intelligence asks: what's happening at the layer above me, below me, and adjacent to me? Is the model layer commoditizing? Is the orchestration layer consolidating? Is the compute layer getting scarcer or more abundant? These questions matter more than anything a direct competitor is doing, because the most lethal threats in the AI era don't come from competitors. They come from layer dynamics.
None of this means the classics have nothing to teach us. Porter's fundamental insight, that competition has structure and that structure can be analyzed, remains true. The BCG matrix's core idea, that portfolio allocation should match the dynamics of each business unit, is still sound. But the specific models, the specific variables they track, and the specific prescriptions they produce are calibrated for a world that no longer exists.
The executives who will make the best decisions in the next decade won't be the ones who can populate a Five Forces diagram most fluently. They'll be the ones who can sense which layer of the stack is shifting, how fast the commodity tide is rising, and whether their position is a quarry or driftwood. That requires not a better framework, but a fundamentally different relationship with uncertainty: one where the goal isn't to achieve the false comfort of a complete analysis, but to maintain the honest discomfort of knowing the territory is changing faster than any map can capture.
Comments
Post a Comment