Start with the mechanism
Going beyond the 7 Powers means testing how competitive advantage changes when AI reduces some costs of building and copying. It does not mean declaring every familiar barrier obsolete. A useful moat explanation still identifies a customer benefit, explains why rivals cannot neutralise it, and supplies evidence for both. The practical change is to examine which part of that barrier depends on work an agent could now perform cheaply.
Hamilton Helmer's framework names seven mechanisms: scale economies, network economies, counter-positioning, switching costs, branding, cornered resources and process power. Those names make a useful shared vocabulary. Their presence in a pitch, however, does not establish that a particular company possesses a Power. A founder needs to describe the relevant competitor, the customer decision and the constraint that keeps the advantage from disappearing.
Our proposed approach is to preserve that vocabulary while opening the underlying mechanisms to challenge. Ask which assets are controlled, which obligations have been met, which outcomes have been observed and which facts a competitor could reproduce. A fresh category earns its place by exposing a barrier the existing categories obscure, rather than by making the same story sound newer.
Separate building from copying
AI capability claims need a task and a date. In an early-2025 randomised study, experienced open-source developers working on familiar mature repositories took longer when AI tools were allowed. That result does not settle the productivity of today's tools; it shows why sweeping statements about every software product being instantly reproducible are poor evidence. A working imitation must meet the original system's relevant quality and operating constraints.
Use a competitor exercise to make that distinction concrete. Write down the visible interface, underlying decisions, integrations, permissions and support obligations. Then ask a technically credible rival to estimate the effort for each part under explicit assumptions. Treat the estimate as a scenario, with uncertainty. A rebuilt screen demonstrates interface reproducibility. It does not by itself demonstrate reliable migration, approved production access or accountability for failures.
This exercise can also reveal that an apparently deep moat is shallow. If the supposed barrier is an undocumented configuration that a customer can export and an agent can reconstruct, the founder should investigate portability. The exercise should be allowed to weaken the claim. A process that only produces flattering answers cannot distinguish an advantage from a persuasive description.
Look for the complements
David Teece's research provides another lens: when imitation is easy, owners of complementary assets may capture the benefits of innovation. This is a theoretical account of value capture, not a forecast for any AI startup. Applied cautiously, it directs attention to what must surround a model or application before a customer can use it successfully: distribution, production access, service capability and other assets the software alone does not supply.
For a founder, the relevant question is who controls those complements. A channel partner that can terminate access immediately offers a different position from a durable, enforceable relationship. A deployed service network is different from a list of prospective installers. Record the scope, conditions and dependence of every important asset. Explain what happens if the platform owner changes terms, a supplier disappears or a customer chooses another operator.
Avoid counting the same fact repeatedly. An exclusive contract might support a resource advantage and help explain customer retention, but those are related consequences of one instrument. Repeating it under several labels should not make the evidence look broader. An explicit causal chain is more useful: controlled access enables a service, the service improves an outcome, and a rival faces a particular obstacle to offering the equivalent.
Treat new moat ideas as hypotheses
Research for the next version of the method explores calendar-time assets, proprietary learning loops, verified accountability and access through agents. These are candidate mechanisms. Their appeal does not establish that they cause lasting advantage. For each, specify an observation that would support it and an observation that would contradict it. A trusted-provider claim, for example, needs customer-choice evidence; a learning claim needs measured improvement against a relevant baseline.
Keep the company-specific evidence separate from the research supporting the general mechanism. An academic paper can show that a mechanism is plausible in a defined setting. It cannot prove that a founder's dataset is exclusive, that customers value its results or that its access rights will survive renewal. Those claims require the company's own documents and outcomes. Evidence should remain traceable to the exact assertion it supports.
A useful review ends with a small proof agenda. Identify the weakest material claim, choose a test that can resolve it, name the owner and record the result whether favourable or unfavourable. This creates a record of learning rather than a collection of moat slogans. It also makes disagreement productive: the reviewer can question a fact or an assumption without arguing about the company's whole identity.
What MoatDefender does today
Today's Moat Index uses a v5.0 method based on the 7 Powers, with tests for AI replication, elapsed time and outcome-based business models. The methodology page explains what the index measures, the evidence it values, its uncertainty and its validation so far. Readings are stable on the same evidence; the method's predictive validity still needs validation. Read the version and limits alongside any number. A Moat Index reading is an evidence-based opinion about defensibility.
A broader next version, the Global Moat Defender Index, graph-backed recipes and company cohort percentiles are coming. They are research and product direction, rather than capabilities implied by this article. Market fit, scalability and sector trends also deserve examination, but blending them into a moat number would hide different questions. A barrier can exist around something customers do not want; demand can grow for something easy to copy.
Use the method to improve the quality of the question: what benefit exists, what prevents neutralisation, and what evidence changes the answer? Better language and a longer feature list are insufficient. The next useful step is usually a document, experiment or observed customer decision that reduces uncertainty about a specific barrier. Nothing here predicts funding, investment returns or a company valuation.
Sources
Sources support the linked claims; our interpretation and checklists establish no outcome for your company.