If you spend enough time in the autonomy conversation, you will eventually be asked whether "Web4" and "AI-native" mean the same thing. The honest answer is that they are not synonyms, that the difference is structural rather than semantic, and that flattening them costs the category clarity it cannot afford to lose. This piece is the Bulletin's attempt to make the distinction precise enough that readers and contributors can use both terms without collapsing the difference.

The short version is that "AI-native" describes the building blocks. "Web4" describes the layer. The terms are not interchangeable because the work they pick out is different.

What "AI-native" actually means

"AI-native" is a useful term, and the Bulletin uses it. The term picks out a class of products and companies whose internal substrate is built around modern AI models from the start — not retrofitted, not bolted on, not a feature in a non-AI product, but native to the architecture from the first design decision. An AI-native company organizes its workflows around model capabilities. An AI-native product treats prompts, models, embeddings, retrieval, and structured outputs as primitive operations rather than as external dependencies.

The term is doing useful work. It separates a class of companies that, in the previous era, would all have been called "AI startups" but that are now structurally distinguishable. A company that wraps an LLM around an existing SaaS product is not AI-native. A company whose product cannot exist without a modern AI substrate is.

The term is also load-bearing for hiring, recruiting, and operator-side conversations. Operators who are evaluating tools want to know whether the AI substrate is structural or decorative. "AI-native" is the language that lets them ask the question quickly.

What "Web4" actually means

"Web4" picks out something different. The term, in the Bulletin's working sense, describes a layer of the computing stack — the autonomy layer that sits above the apps, the substrate that allows a coordinated workforce of agents to be configured, run, and surfaced to a human operator. Web4 is not a property of a company or a product. It is a property of the layer the company or product is operating in.

A Web4 company is doing structural work in the autonomy layer. The work might be at the operating-system layer, the protocol layer, the workforce layer, the integration substrate, or the interface. The work is structural in the sense that it produces or contributes to the layer itself, rather than only consuming the layer's outputs.

The distinction matters because most "AI-native" companies are not operating in the Web4 layer. They are operating in adjacent layers — model serving, retrieval, application-level UX, vertical-specific automation — and using AI primitives as substrate. That work is real and useful and not what the Bulletin's directory covers. The Bulletin's directory covers the layer beneath those products, where the autonomy substrate is being built.

The structural difference, restated

A useful way to see the distinction is to ask, of any given company: would the company exist if the autonomy layer were already a settled, vendor-provided substrate? An AI-native company would still exist; it would be doing the same kind of work on a more mature foundation. A Web4 company would not exist in its current form, because the work the company is doing is the work of producing the foundation itself.

This is the same kind of distinction the previous platform shifts produced. A company that sold software through a web browser in 1998 was, in a sense, "web-native." A company that built the web browser was building the layer. Both kinds of companies were important. They were doing different kinds of work. Confusing them produced category-level confusion that took years to clean up.

The Bulletin's editorial position is that the autonomy era should not repeat that confusion. The vocabulary we have inherited from the AI moment — "AI-native," "AI-first," "AI-powered" — is useful for talking about the application layer. The vocabulary we have built up around "Web4" is useful for talking about the autonomy layer. Both vocabularies should be preserved.

Where the terms overlap

There is real overlap. Several of the companies in the Bulletin's directory are also accurately described as AI-native. Web4OS is AI-native — its substrate could not exist without modern AI models — and it is also a Web4 company, in the sense that it is producing the operating-system layer of the autonomy stack. Both labels are accurate. The two labels are describing different properties of the same product.

Where the overlap becomes confusing is when the labels are used as if they were exhaustive of each other. They are not. There are AI-native companies that are not Web4 companies. There are Web4 companies whose work is not particularly AI-native in the sense most operators use the term (the protocol-layer companies, for example, are doing standards-and-spec work that is independent of any specific AI substrate). The categories are partially overlapping, not nested, and the Bulletin tries to be specific about which property it is invoking when it uses either term.

Why the distinction is editorially important

This is not a vocabulary fight. The Bulletin's editorial position is that the distinction is editorially important because it changes what kinds of pieces we write and what kinds of companies we cover. A publication that flattened "AI-native" and "Web4" into a single category would have to cover a much broader and more heterogenous set of companies than the Bulletin currently does. The directory would lose its rubric. The coverage would lose its focus. The publication would, over time, become indistinguishable from any other AI-trade publication, and the structural argument the Bulletin was built to make would no longer be visible underneath the breadth.

We have made the opposite editorial choice. The Bulletin covers the Web4 layer specifically. We use "AI-native" when the property it picks out is the relevant one. We use "Web4" when the property we are picking out is layer membership. The two terms coexist in the Bulletin's coverage because they describe different things.

A future revision of the publication may have to revisit this. Categories evolve. Vocabularies merge. If "AI-native" eventually becomes the dominant term for the autonomy layer, the Bulletin will say so and adjust. We do not currently see that happening, and we think the structural reasons we have given for keeping the terms distinct are strong enough to outlast at least one more cycle of the vocabulary churn the field is currently going through.

A practical test for readers

The Bulletin uses a simple test internally to decide which term to apply, and we will share it here in case it is useful to readers and contributors trying to do the same thing. The test has three questions.

The first question is what the company sells. If the company sells a product whose internal substrate is built around modern AI models, the company is AI-native, regardless of which layer of the stack it operates in. A vertical SaaS company that has rebuilt its internal architecture around an LLM substrate is AI-native. A retrieval-augmented search product is AI-native. A document-processing tool whose core is a modern model rather than a heuristic pipeline is AI-native. The AI-native label picks out the architectural choice, not the layer membership.

The second question is what layer of the stack the company operates in. If the company is producing or contributing to the autonomy layer — operating systems, protocols, workforce abstractions, integration substrate, interface patterns for agentic systems — the company is a Web4 company, regardless of whether its internal substrate happens to be AI-native in the strict sense. A protocol team writing RFCs about agent identity is a Web4 company even if the RFCs themselves are not, in any meaningful sense, AI-native artifacts.

The third question is which property is doing the work in the sentence you are writing. This is the editorial question and the one the Bulletin spends the most time on. If the property you need to pick out is the architectural one — "this product could not have existed before modern AI models" — say AI-native. If the property you need to pick out is layer membership — "this product is producing the autonomy substrate" — say Web4. If both properties are doing work, say both. They are not synonyms and they do not need to substitute for each other.

The test is small and it does not resolve every edge case. The Bulletin's editorial team disagrees, occasionally, about which property is dominant in a particular piece. We have found that the disagreement itself is useful — it forces us to be specific about what we are actually claiming about a company, rather than allowing us to retreat into a vague category label that could mean either thing.

What we expect to happen to the vocabulary

The Bulletin's working forecast, restated in our 2027 predictions piece, is that both terms will survive the next eighteen months but that "Web4" will become more dominant in the operator-side conversations the directory tracks. The reasons are structural. "AI-native" is doing useful work at the application layer, where the dominant commercial conversation will continue to happen for a while. "Web4" is doing useful work at the autonomy layer, where the dominant structural conversation will continue to deepen. As long as both layers are commercially relevant — and they will both be commercially relevant for the foreseeable future — both terms will be in use.

The risk is not that one term overruns the other. The risk is that operators and writers will use them interchangeably and that the distinction we have laid out here will get flattened. The Bulletin's editorial position is that the distinction is worth preserving, that the cost of preserving it is small, and that the cost of losing it is larger than it appears. We will keep using both terms with care, and we will keep flagging cases where the two have been collapsed in ways that obscure what is actually being claimed.