Anette Mellbye's Post

Anette Mellbye · article · 2026-04-20 source ↗ Reader 🖥️

Moats & Defensibility Riding Model Improvements

AI companies must determine early whether they are building durable infrastructure — characterized by switching costs, customer unlearning, and accumulated institutional knowledge — or mere wrappers that platform providers will absorb in their next release.

practical · argumentative — A practical argument aimed at AI company builders, providing a diagnostic framework (the 'infrastructure test') to evaluate whether their product is durable or a vulnerable wrapper. Uses concrete examples (Chegg, Consigli, Strise, Figma) to support each point.

Structure

Part	Role
Opening provocation	Frames the core strategic question: what remains of your company when the platform absorbs your function?
Definition of wrappers (via Trond Riiber Knudsen)	Introduces the concept of a 'wrapper' — a company layered on top of others' capabilities with no defensible core.
Speed-of-disruption claim	Argues that LLMs compress the timeline for competitive displacement from quarters to releases.
The four-question infrastructure test	Core framework: switching costs, customer unlearning, usage-driven improvement, free-feature risk.
Chegg as cautionary tale	Negative example — a company destroyed because its offering was easily replicated by a platform.
Consigli and Strise as positive examples	Companies that pass the infrastructure test due to accumulated institutional knowledge and high switching costs.
Figma as ambiguous case	Shows the difficulty of determining infrastructure vs wrapper status even for established companies.
Privileging question 2 (unlearning)	Argues that the most important test is whether the customer has lost the ability to perform the function themselves.
Call to action for Norwegian AI companies	Concluding normative claim: ask the four questions early to build something lasting.

Key terms

Wrapper (wrapper) Moats & Defensibility
A company that has not built anything proprietary, merely layering on top of a platform it does not own — vulnerable to absorption by the platform provider.
Infrastructure test (infrastrukturtest) Moats & Defensibility
A four-question diagnostic to determine whether a product is durable infrastructure or a vulnerable wrapper.
Unlearning (avlæring) Moats & Defensibility
The process by which a customer loses the internal capability to perform a function themselves, making them dependent on the provider regardless of switching costs.
Switching costs (byttekostnader, implied) Moats & Defensibility
The practical effort required for a customer to replace one solution with another — necessary but not sufficient for durability.
Institutional knowledge (institusjonell kunnskap) Moats & Defensibility
Organization-specific accumulated decisions, risk assessments, and compliance history that belong to the customer and cannot be transferred to a shared model.

Propositions

explicitMost AI companies start as wrappers — layered on platforms they do not own.conf: high
explicitdevelopedPlatform providers (OpenAI, Anthropic) will absorb wrapper functionality into their own products.conf: high
extends ←Is Your AI Training the Competition? · p2 — Generalizes the 'platform absorbs wrapper functionality' dynamic (there, about B2B SaaS wrapper products) to a new target: any enterprise customer's internal problem-solving expertise, absorbed through ordinary use rather than feature-copying.
extends ←This AI notetaker won't sell surveillance to your boss · p10 — Newton names a specific mechanism (MCP connectors turning Granola into a headless memory layer for Claude/ChatGPT) by which a platform could absorb a wrapper's value, concretizing Mellbye's general claim that platform providers will absorb wrapper functionality.
explicitLLMs have compressed the timeline for competitive displacement from quarterly cycles to release cycles.conf: high
explicitCustomer unlearning (loss of internal capability) is a stronger moat than high switching costs.conf: high
explicitThe four infrastructure questions must be asked before the outcome is determined, not after.conf: high
implicitcorroboratedA company that accumulates customer-specific institutional knowledge becomes irreplaceable regardless of platform evolution.conf: medium
supports ←This AI notetaker won't sell surveillance to your boss · p9 — Pedregal's own moat argument for Granola (accumulated, agent-ready personal context as the defensible asset, not the notetaking feature) matches Mellbye's thesis that accumulated customer-specific institutional knowledge, not the wrapper feature, is what becomes irreplaceable.
implicitMost Norwegian AI companies have not asked these four questions early enough.conf: medium
explicitEven established companies like Figma cannot be certain whether they are infrastructure or wrapper.conf: high

Arguments

AI companies must determine early whether they are infrastructure or wrappers, because platform absorption is inevitable and fast. strength: high

Explicit premises:

Most AI companies start as wrappers on platforms they don't own.
Platform providers will absorb wrapper functionality in future releases.
The speed of displacement has accelerated from quarters to releases with LLMs.

Implicit premises:

Being absorbed by a platform means the wrapper company loses its reason to exist.
Early self-awareness about wrapper status allows a company to pivot toward infrastructure.

Inferences:

The four-question infrastructure test should be applied as early as possible in a company's lifecycle.

Customer unlearning is a more durable moat than switching costs. strength: high

Explicit premises:

A customer with high switching costs can still switch.
A customer who has unlearned has lost the ability to choose — independent of whether they know it.

Implicit premises:

Switching costs are quantifiable and can be overcome with sufficient motivation or a sufficiently better alternative.
Unlearning is a cognitive/organizational change that is qualitatively different from a cost barrier.

Inferences:

Question 2 ('Has the customer stopped doing this job themselves?') is the most important of the four infrastructure questions.

Chegg was destroyed because it was a wrapper, not infrastructure. strength: high

Explicit premises:

Chegg had 19 million customers.
When ChatGPT launched, Chegg's stock fell ~48% in one day.
Chegg's core offering suddenly existed elsewhere — cheaper, faster, simpler.
Chegg itself warned investors about AI pressure.

Implicit premises:

Chegg's homework help did not create customer unlearning — students could easily switch to ChatGPT.
Chegg had not accumulated irreplaceable institutional knowledge.

Inferences:

Chegg would have failed all four infrastructure-test questions.

Narratives

Chegg had 19 million customers but lost 48% of its stock value in a single day when ChatGPT offered equivalent homework help cheaper and faster.
→ Cautionary tale illustrating the existential risk of being a wrapper.
Strise accumulates bank-specific compliance decisions, risk assessments, and resolved cases over time, creating institutional knowledge that belongs to the bank alone. Switching would mean starting over while the regulator watches.
→ Positive example of infrastructure: demonstrates unlearning and accumulated institutional knowledge as moats.
Figma's stock fell 7% when Claude Design launched, revealing market uncertainty about whether Figma is infrastructure or wrapper.
→ Illustrates the ambiguity of the wrapper/infrastructure distinction even for mature companies.

Questions addressed

✅ How can an AI company determine whether it is building something durable or merely a wrapper?
Apply the four-question infrastructure test: switching costs, customer unlearning, usage-driven improvement, and free-feature risk.
✅ Why are AI wrappers more vulnerable now than analogous companies were in previous technology waves?
LLMs compress the displacement timeline from quarterly earnings cycles to software release cycles.
✅ Which of the four infrastructure questions matters most?
Question 2 (customer unlearning) — it describes an irreversible change in the customer, not just a cost barrier.
✅ What does it look like when an AI company is genuinely infrastructure?
Examples: Consigli (passes all four) and Strise (accumulates bank-specific institutional knowledge).

Gaps

incompleteThe four-question framework is presented without evidence of its predictive validity. It is asserted as useful but not tested against historical cases beyond the selected examples.
Read next: Compare with established frameworks for competitive moats (e.g., Hamilton Helmer's 7 Powers, or Christensen's disruption theory) to evaluate whether the four questions add genuine diagnostic power.
incompleteThe post does not address the possibility that wrappers can deliberately evolve into infrastructure — the transition path is left unexplored. The binary framing (wrapper vs infrastructure) may be too rigid.
Read next: Look for case studies of companies that started as wrappers and successfully built infrastructure moats over time.
uninformedThe Figma example conflates stock-price reaction with actual strategic vulnerability. A 7% drop on a competitor announcement is not evidence that Figma is a wrapper; it may simply reflect short-term market uncertainty.
Read next: Examine Figma's actual switching costs, network effects, and design-system integrations to make an independent assessment.
incompleteThe concept of 'unlearning' is introduced as the strongest moat but is not analyzed for its ethical implications — is it desirable for vendors to engineer customer dependency through capability atrophy?
Read next: Consider the ethics of vendor lock-in through deliberate or incidental deskilling of customer organizations.