A former iPad design app has become an AI campaign platform, and its pivot says more about the startup industry than it does about design.
Linearity used to be easy to explain. Previously known as Vectornator, it was a relatively accessible vector-design app built largely around the iPad and Mac. It offered an alternative for people who wanted to create illustrations and graphics without committing to Adobe’s larger and more complicated ecosystem.
Today, Linearity describes itself as an AI design generator that can turn a prompt into a collection of editable, on-brand campaign assets. The vector editor is still there, but it is no longer the main story. The company now presents AI-assisted campaign production as the centre of the product. And the reviews aren’t great.
It is difficult not to see the commercial logic behind the change. A pleasant iPad design app belongs to a crowded software category with established competitors and limited room for dramatic growth. An AI platform for enterprise marketing teams can be sold as something larger, more urgent and considerably more expensive.
Linearity is hardly alone. This is where much of the software industry is heading.
A large number of startups now sit between users and a small group of foundation-model companies. They take technology from OpenAI, Anthropic, Google, Stability AI or another provider, add an interface and build a product around one particular task. The result might write advertisements, analyse contracts, create presentations, generate product photographs or answer questions about uploaded documents.
These products are usually described as AI companies, even when the underlying intelligence comes mostly from somewhere else.
Linearity’s own AI terms state that some of its tools use technology supplied by third-party providers and specifically identify Stability AI as one of those providers. That does not mean Linearity has built nothing, but it does place the product within the wider application layer being constructed on top of foundation models.
The word commonly used for these companies is “wrapper,” and it is usually intended as an insult. It suggests that the company has done little more than place a polished interface around an API.
That description is sometimes fair, but it can also be lazy. Almost all software depends on technology created by somebody else. A startup does not need to train a language model or build its own cloud infrastructure to create a worthwhile product. Making difficult technology understandable, reliable and useful is a legitimate form of product development.
A good wrapper can understand a particular profession better than a general chatbot. It can connect the model to company data, existing software, approval processes and industry-specific rules. It can provide security, support and predictable output. For many customers, those details are more important than knowing who trained the model.
The problem is not that wrappers exist. The problem is that many of them are extremely thin.
A prompt box, a collection of preset instructions and a subscription page may be enough to launch a product, but it does not necessarily create a company with a lasting advantage. The underlying model providers are moving quickly, and every improvement they make can remove the reason for several smaller products to exist.
Claude Design is a useful example of that pressure. Anthropic introduced it as a product for creating designs, prototypes, slides and other visual work, with the ability to move results into tools such as Canva. Claude already works across writing, documents, research and code, so design can become another part of a broader working environment rather than a separate AI subscription.
That does not mean Claude Design will automatically produce better work than Linearity. A specialist design application can offer finer controls, more reliable editing and a workflow developed around the needs of designers. General models still produce plenty of awkward layouts and questionable creative decisions.
It does, however, expose the risk facing the wrapper economy. When the company supplying the underlying model begins moving into your category, the wrapper has to prove that its additional layer is genuinely valuable.
For Linearity, that value could come from editable vector output, brand systems, resizing, translation and the existing Curve and Move applications. Its website places particular emphasis on generating complete campaigns that remain editable rather than producing flat images. Those are practical distinctions, especially for marketing teams that need to turn one idea into assets for several channels.
Still, there is a large gap between making campaign production more convenient and changing the nature of creative work.
Most companies do not suffer from a shortage of marketing content. They already publish more banners, emails, social posts and landing pages than their audiences can reasonably care about. The harder problem is deciding what is worth saying, developing an idea that deserves attention and getting an organisation to agree on a clear message.
Generating more variations does not solve that problem. It may simply allow an average idea to spread across more formats at greater speed.
This is one of the stranger contradictions in the current AI market. The technology is regularly presented as a way to remove unnecessary work, while many products are designed to generate enormous quantities of material that someone will still need to inspect, approve, distribute and measure.
The startup incentives are understandable. Building a model is expensive, while building with an API has become relatively accessible. Investors and customers are also more interested in AI than they are in another conventional SaaS product. An existing company therefore has every reason to rewrite its story around AI, even when the underlying product has changed less than the branding suggests.
Linearity is a particularly visible example because the transition is so clear. A company once associated with a focused Apple design app now talks about automated, on-brand campaign production across teams and markets. The software may be useful, but the repositioning also reflects the pressure on every technology company to present itself as part of the AI economy.
The more important question is whether the business has created anything the model providers or larger software platforms cannot easily reproduce.
The wrappers that survive will probably be the ones that own more than the interface. They will have specialist data, deep integrations, trusted customer relationships and an understanding of workflows that cannot be recreated by adding another feature to Claude or ChatGPT. Their products may still depend on outside models, but changing the model underneath them will not destroy the reason customers use them.
The thinner products will face a less comfortable future. Their main features will gradually appear inside larger platforms, model prices will change, customers will become tired of maintaining separate AI subscriptions, and the novelty of placing “AI” beside a company name will disappear.
Linearity AI may eventually prove that its design workflow is strong enough to justify its place. It may also become another example of a software company following the market, attaching itself to technology developed elsewhere and hoping its position between the model and the customer remains valuable.
That is not necessarily dishonest, and it is not unique to Linearity. It is simply the business model behind a growing part of the AI startup industry.
The current generation of wrappers is helping people understand what foundation models can be used for. The next stage will be less forgiving. Once the excitement settles, these companies will no longer be judged by whether they contain AI. They will be judged by whether there is still a reason for the company itself to exist.
