Long Live the Wrapper
The line between consumer companies and infrastructure companies is beginning to blur.
For the better part of the last decade, there has been a fairly consistent view of technology: value accrues to the deepest layer.
Infrastructure mattered more than applications. Data mattered more than interface. And now, in the age of AI, models are assumed to matter more than everything else (except for maybe chips). The logic is clean. The closer you are to the source of intelligence or the ‘hard thing’ the more control you have — and the more value you capture.
Which is why the industry has begun describing these application-layer companies, often dismissively, as “wrappers.” These are “thin” experiences sitting on top of OpenAI, Gemini, or Anthropic that create a particular experience or outcome for a user. The assumption has been that they could easily be absorbed by the models themselves. And that that they are ephemeral because their functionality can be simple to duplicate (unlike the models).
And yet, if you zoom out, that framing feels incomplete.
Because the most enduring consumer companies have never really been defined by the depth of their technology. No one looks at the supremacy of TikTok, Airbnb, Meta, Uber, etc. and points to novel technology breakthroughs that these companies owned. They borrowed the best technology of their time and created novel, sticky experiences that met a human need. Their success has been defined by something harder to measure and define than a deep technology breakthrough, and often only visible in hindsight.
They’ve been defined by taste.
Taste As An Input
There’s a tendency to treat taste as surface-level: as branding, or aesthetics, or some vague sense of design polish. But that interpretation misses where taste actually lives inside a company.
Taste isn’t what a product looks like. It’s how decisions get made.
It shows up in what gets built — and, more importantly, what doesn’t. It’s present in the features that are left on the table, the edges that are intentionally smoothed over, the moments where a team chooses restraint over expansion. It’s coherence.
In that sense, taste isn’t an output of the system. It’s an input, a lens through which everything else is filtered.
For a long time, that distinction didn’t feel top of mind. Constraints did most of the work. It was hard enough to build anything at all, so the question of what should be built was often secondary to what could be built. A founder friend describes this in his benefit-to-cost analysis.
AI changes that equation.
From Creation to Curation
What AI has done, perhaps more than anything else, is expand the space of possibility.
It is increasingly trivial to generate text, images, code, even entire products. The cost of creation is collapsing, and with it, many of the traditional bottlenecks that once defined software development.
But as the space of possibility expands, a new constraint emerges. If everything can be built, what is actually worth building?
This is where the conversation around “wrappers” starts to break down. Because it assumes that the primary challenge is still execution — that the hardest part is generating the output.
The challenge is selection. It’s knowing which of the infinite possible directions to pursue, and which to ignore. And that is a fundamentally different problem — one that models, by themselves, are not particularly well-suited to solve (today at least).
Model companies optimize for accuracy, latency, and generalization. Wrappers can instead optimize for emotion, identity, and cultural resonance — whether through community, personalization, or just pure joy.
When done well, wrappers feel more like filters. Or curators. They own the relationship and that’s where lasting value can stack up. Defensibility emerges by avoiding becoming a ‘feature’ and instead becoming the default destination people trust. Loyalty and emotional stake is hard for model providers to produce at scale.
Wrappers win when they:
Mediate identity, not just output
Shape taste, not just answers
Create participation, not just automation
“Bad” wrappers are thin UI on top of LLMs that lean into pure utility. We do see a lot of these. More interesting wrappers accumulate user-specific context over time, encode preference graphs, create participatory loops, and build social or cultural layers.
Think about the companies we flagged earlier:
Uber: not better maps, but trust + reliability UX
Airbnb: not better listings, but belonging + narrative
TikTok: not just ML, but taste formation engine
And all this is powered by taste.
Wrappers, Reconsidered
None of this is to say that models or infrastructure don’t matter. They do. In many cases, they define the baseline of what’s possible. Apps wouldn’t exist without cloud computing and smartphones, for example.
But as that baseline rises — as intelligence becomes more accessible, more abundant — the locus of differentiation begins to shift away from raw capability, and toward the choices made on top of it.
The question is no longer just who can build the most powerful system. It’s who can decide what that system should actually do.
The Collapse of Layers
At the same time, there’s another shift happening: the line between consumer companies and infrastructure companies is beginning to blur.
Historically, these were distinct paths. Consumer companies focused on experience and distribution. Infrastructure companies focused on capability and scale. The former owned the relationship, the later owned the system.
But in this new world, the two are becoming increasingly intertwined. The experience is no longer just a layer on top of the system; it is part of the system. User interactions generate data, that data improves the underlying models or workflows, and those improvements, in turn, reshape the experience.
It becomes a loop that is difficult to disentangle.
There are early companies pioneering this new iteration (beyond OpenAI and Anthropic as poster children):
Rythm - At a glance, Rythm looks like a consumer diagnostics company — helping individuals better understand and navigate their health. But beneath the surface, the company is also building the logistics, lab integrations, and regulatory scaffolding to deliver at-home diagnostics to other businesses. A consumer surface, underpinned by owned infrastructure.
Nilo - Nilo is a consumer play platform or AI-native space for creating and exploring interactive experiences. But it is also the tooling and orchestration layer that enables those experiences to exist at all. The experience is the entry point that invites participation; the infrastructure enables it. Over time, that loop compounds, where taste shapes creation, and creation feeds the system.
ElevenLabs - The value proposition of ElevenLabs is deceptively simple. To creators and developers, it’s a generator of high-quality, realistic voice. At the same time, they are building the core models, tooling, and infrastructure that define how voice is created, controlled, and deployed across applications. A surface for developers and creators, but buoyed by proprietary infrastructure.
Closing Thoughts
As AI continues to improve, the cost of creation will continue to fall. But the cost of judgment will not.
Because knowing what to build — what matters, what resonates, what endures — is not just a function of more intelligence.
It’s a function of taste. And in a world where almost anything can be generated, that may be the only thing that truly cannot.
