The company DreamStories publishes children’s books. His stories do not stand out for their originality – a boy who travels by rocket to a planet inhabited by dinosaurs – but they have a distinctive feature: the protagonist is the reader himself, converted into a comic character. This personalization, which turns each book into a unique piece, is not signed by an illustrator or a writer, but by a generative artificial intelligence (AI) model that produces text and images on demand. The case of this start-up, founded in Madrid by Phil Calzavara in 2023, is not isolated. Vidext uses this technology to transform corporate documents into multimedia content. Vibepeak uses it to create advertisements and Aithor uses it to generate academic texts. The list is long. As Calzavara summarizes: “When a new technology emerges, there are many people who start looking for business.” Although the idea seems brilliant—AI does the job faster and, above all, cheaper—the future of many of these companies is uncertain. Experts point out two risks: above, that large technology companies integrate these services into their own models; and below, too simple proposals are replicated by the competition and diluted in a saturated market. For David Gordo, professor at IE Business School in Madrid, these start-ups fulfill a useful function: they take an advanced technology and bring it down to earth to commercial use cases. “The leap to go from the general to the concrete is very difficult,” he maintains, and that is why he believes that “some companies are going to remain,” but warns that “a large majority will not.” The danger is the same as always: that the big fish eats the little one. Gordo explains that, if the service they offer has a large market, giants like OpenAI, Google or Anthropic can launch their own version and keep the business. There are already examples of this: the companies Lovable and Cursor, programming assistants, now compete with Google Antigravity, the product that the technology giant launched in that segment in November. AI-assisted programming has a huge market: according to a study by JetBrains, a developer of tools for writing code, 85% of programmers use AI tools in their work.
Generalist models
Many of the companies that have appeared with this new technological wave do not develop AI models from scratch. They rely on large generalist models—such as GPT (created by OpenAI), Gemini (Google) or Claude (Anthropic)—and build their image, text or video generation system on top of them. Its main value consists of transforming the generalist capacity of the models into a concrete service through precise instructions and the design of interfaces that allow users to interact with the technology. In sector jargon, these instructions are known as prompts: more or less elaborate texts that explain to the system what it should produce, with what style or format. Writing prompts is, in many cases, the operational core of these companies, which are known as IA wrappers, a term that translates as AI wrappers. Esteve Almirall, professor of Innovation at the Esade University of Barcelona, points out that AI wrappers have been the first commercial version of generative AI. He emphasizes that they are not a Spanish phenomenon; It has happened all over the world. And he warns of the second great risk: “If the technological value they provide is very low, 20,000 rivals will copy it immediately.” Gordo agrees: “If it is very easy to imitate, others will do it.” The two professors share the opinion that the border between an innovative product and an imitable one is constantly moving because technology advances at breakneck speed. It is precisely that border that Kfund analysts try to draw. Jaime Novoa, partner of the investment firm specialized in technological ventures, explains that they focus on start-ups aimed at very specific industries. Its strategy consists of identifying projects that combine expert knowledge in a specific sector—such as cybersecurity or gaming—with generative AI capabilities. In his opinion, it is unlikely that “horizontal” platforms such as OpenAI or Google will compete in these “vertical” markets. But he warns, “it is not impossible that they end up doing it.” When evaluating the possible financing of a business, Kfund asks themselves a question: “How difficult is it for a third party to set up something similar?” When entry barriers are low, Novoa thinks that a start-up that develops a product in Spain has a good chance of encountering competitors that do the same in larger markets, such as the US. In this context of global competition, the analyst highlights the qualitative leap in Spanish technological projects in 2025, both in quantity and quality: “We see start-ups with a level of ambition and technique that we had not seen before.” Calzavara, founder of DreamStories, has had competitors appear everywhere. Small and large; in Spain and abroad. Even Google has created a similar product with Gemini Storybook. He recognizes that the technology he uses is not a competitive advantage: “Everyone has access to it,” he declares. The differential factor in the quality of a generative AI product, according to the entrepreneur, is marked by the ability to get the most out of the technology. In its case, it combines different general models to build what it calls its “own generation engine.” Then, it specializes it with context: it feeds it with data and precise information that the AI must always remember when generating the product. At this point it is about providing the system with memory. All this work, or training, as it is called in technological jargon, allows the generation engine to respond in a more refined way in each new line of text, illustration or character that it creates for the books that the start-up sells.
