The competitive advantages that protected businesses for decades are eroding faster than most executives realize, and the replacements look nothing like the original.
Warren Buffett's moat has a leak. So does yours.
The most dangerous position in business right now is not ignorance about AI. It is confidence. The confidence of an executive who has assessed their competitive position, concluded the moat is solid, and gone back to optimizing the business they already have. That confidence is being quietly undermined by a shift that does not show up in quarterly results until it is too late to respond.
This is not incremental disruption. When new technology nibbles at the edges of existing business models, smart companies adapt and defend. That is not what is happening now. The structural logic of competitive advantage itself is being rewritten. The companies that lose the next decade will not be the ones who ignored AI. They will be the ones who added AI to a strategy designed for a different era.
Before dismantling the consensus, it is worth respecting it. Buffett and Munger did not build Berkshire's record by being sloppy thinkers. The moat framework is genuinely useful. A moat is a durable competitive advantage that protects a business from competition over time. The wider the moat, the harder it is for rivals to erode your returns. For decades, moats fell into five reliable categories.
The friction and expense a customer absorbs to move to a competitor, whether data migration, staff retraining, process re-engineering, or contract penalties.
The product becomes more valuable as more people use it, compounding a scale advantage challengers cannot easily replicate.
Scale, proprietary processes, or structural efficiencies that let a company produce or deliver at lower cost than anyone else.
Brands that command pricing power, patents that block competition, regulatory licenses that restrict entry.
Markets where one or two incumbents satisfy all existing demand, making new entry economically irrational.
This framework worked because these advantages were slow to build and slow to erode. Capital, time, and complexity acted as natural barriers. The model assumed a relatively stable rate of technological change. For most of the 20th century and much of the 21st, that assumption held.
It no longer does.
Start with switching costs. The "too embedded to switch" argument used to be bulletproof. Your data lives here. Your people are trained on this. Your workflows are built around us. AI agents now learn new tools rapidly, abstracting away the UI and workflow friction that created stickiness. Data portability, once a genuine technical nightmare, is becoming automated. AI can migrate, reformat, and re-integrate data at costs unthinkable five years ago. The mid-market SaaS companies most exposed are the ones whose retention metrics look strong precisely because switching friction is high, not because customers love the product. That distinction is starting to matter.
Scale-based cost advantages are dissolving. AI collapses the labor and expertise cost curves that scale once depended on. A 10-person company can now execute what previously required 500: customer support, content production, legal review, code generation, data analysis. Operational complexity, long a barrier to entry, is being automated away. Scale still matters for capital-intensive infrastructure. For knowledge work and services, the advantage is evaporating.
Brand and IP face a different kind of pressure. AI-generated content floods every channel, making differentiation through content production nearly impossible. Patents are harder to defend when AI can design around them faster than legal processes can respond. Companies relying on brand as a primary moat need to understand what brand actually means in this environment. Not aesthetics. Trust.
Network effects remain the most durable surviving moat. They are not immune. AI enables faster bootstrapping of competing networks. New entrants can simulate early-stage engagement and accelerate adoption curves in ways that were not available before. The moat holds where the network creates irreplaceable relationships and coordination. It does not hold where the "network effect" was mostly data accumulation dressed up in better language.
Efficient scale qualifies fewer markets than it used to. The assumption that a given market can support only one or two players is being tested in category after category. Infrastructure-level scale remains a real barrier. Application-layer scale advantages are shrinking fast.
Most companies are not ignoring AI. They are using it. Adding AI features to existing products, deploying AI tools to cut costs, announcing AI strategies in earnings calls. What they are not doing is asking the right question.
The question most companies ask:
"How do we use AI?"
The question that actually matters:
"What does our competitive position look like if AI erases our current advantages?"
Companies asking the second question are building differently.
Three mistakes appear most often. The first is confusing product differentiation with durable advantage. Product capabilities that took years to build can now be replicated in months. Feature parity is table stakes. The companies most confident in their product moats are the most exposed, because they have the most to lose and the least urgency to act.
The second is defending yesterday's perimeter. Capital flows into sustaining existing moats that are already eroding. The logic is not irrational in the short term. A business with $200M in ARR, strong renewal rates, and an installed base that is not actively churning has every financial incentive to protect that position. The problem is that the churn eventually comes, and by the time it shows up in the metrics, the window to build something new has typically already closed.
The third is underestimating the speed of capability diffusion. Historically, new capabilities took years to spread through markets. AI capabilities are diffusing in months, sometimes weeks. Any competitive position built on "we have this capability and our competitors don't" has a shelf life measured in quarters, not years.
Five categories of moat are strengthening in an AI-native competitive environment.
1Proprietary data loops, not just data. Owning data is not a moat. Every company claims data as an asset. A moat is a flywheel: using your product generates better data, which improves the product, which attracts more usage, which generates better data. Consider a clinical documentation platform where every physician interaction trains models that improve coding accuracy. Competitors cannot replicate that flywheel without the patient volume, and the patient volume requires trust that only comes with time. The question is not "do we have data?" It is "does our system get smarter faster than our competitors' systems?"
2Embedded workflows and behavior change. If your product has fundamentally changed how people work, not just automated a task but reshaped the entire workflow around itself, that stickiness runs deeper than any feature. This is switching costs 2.0: not "it is hard to move our data" but "we rebuilt our operating model around this and there is no obvious path back." AI makes surface-level workflow embedding fragile. Deep operational integration, the kind where removing the product means rebuilding how work actually gets done, is where real durability lives.
3Trust in high-stakes categories. In health, finance, legal, and enterprise, humans still gate consequential decisions on trust. AI makes the trust deficit larger, not smaller. As AI-generated outputs proliferate, verified, accountable, relationship-backed judgment becomes more valuable. Companies that have built genuine trust with specific buyer personas in high-stakes categories own something AI cannot manufacture and incumbents cannot easily transfer.
4Distribution and relationship ownership. Who already has the customer's attention and their buying relationship? Direct relationships with decision-makers compound in value as the cost of reaching buyers through content and advertising increases. Owned distribution is a moat that AI strengthens rather than erodes, because AI raises the noise floor everywhere else.
5Organizational learning velocity. The companies that absorb, deploy, and iterate on new AI capabilities faster than competitors build a compounding structural advantage. The signal is not "we have an AI strategy." It is whether the company can spot a new capability, run a real pilot in four weeks, make a kill-or-scale decision in eight, and have it in production before competitors have finished their vendor evaluation. That speed is organizational, not technological. It comes from how decisions get made, not which tools are licensed.
Write down your current moat. Be specific. Then run it through each of the erosion mechanisms above. What survives?
For most companies, the honest answer is: less than they think. That is not a reason for panic. It is a reason for precision. Identify which of the new moat categories you can credibly build, and start compounding now.
The companies that win the next decade will not be the ones who had the strongest moats in 2020. They will be the ones who recognized the shift earliest and built the right defenses before they needed them.
The best time to audit your moat was two years ago. The second best time is today.
A conversation with one of our Growth Architects is a good place to start. They work through exactly this kind of audit with leadership teams: what you have, what is actually holding, and where to build next.
