An ongoing study of how early-stage founders use generative AI across the whole innovation process — idea, concept, prototype, validation, business model — rather than in the build alone. Interviews with founders in the Oulu region, run through Oulu University of Applied Sciences. Fieldwork starts in November 2026; there are no findings yet, and this page will say so until there are.
Building a working product used to take a technical founder or a team, and a build–measure–learn cycle took months. Generative AI has cut that to days. In Y Combinator's spring 2025 batch, around a quarter of the startups had codebases that were almost entirely AI-generated (TechCrunch, 6 March 2025).
The interesting part is not the speed. It is what the speed does to everything either side of it. If building becomes the cheapest thing a founder can do, it also becomes the easiest way to avoid the expensive question — whether anyone wants the thing. A process that used to force a founder to talk to customers because building was too costly to risk no longer forces anything.
So the question is which parts of the innovation process genuinely changed and which only got faster. Nobody has good evidence on that yet, and founders, ecosystem organisations, and investors are all making decisions as if they do.
How do founders of early-stage startups describe their use of generative AI across the phases of their innovation process?
Three sub-questions narrow it:
AI-assisted development has already been studied at Finnish universities of applied sciences. Several 2026 theses build a minimum viable product with AI tooling and document the process in detail, and one interviews software professionals about the conditions for using AI professionally. Their shared result is that the cost of building has collapsed and the developer's role moves from producing code to judging it.
Those are computing theses. They look at one phase of the innovation process — the build — and four of the five are the author describing their own project. What is missing is what happens to the other phases and to the relationship between them: whether the founder's work moves into validation, or whether cheap building lets them skip it.
The peer-reviewed literature on generative AI in entrepreneurship is thin because the phenomenon is three years old. There is a systematic review of 83 articles, a framework paper on AI in entrepreneurial education and practice, and a study of AI-assisted growth strategies in startups. Some of the most-cited material is preprints. News coverage describes the scale of the phenomenon but cannot carry a theory base, and this study treats it accordingly.
The study is qualitative. There is no established measurement instrument for this, and the useful material is in how founders account for their own decisions — which is what interviews get at and a survey does not.
The material is six to eight semi-structured interviews with founders in the Oulu region who have used generative AI in developing their product. Participants are selected purposively through the Founders of Oulu network, business services, and existing contacts; the final number depends on when new interviews stop producing new themes.
The interview themes come out of the theory rather than out of curiosity: ideation and concept work, prototyping and AI tooling, customer validation and product–market fit, and business model development. Interviews are recorded, transcribed, and analysed with qualitative content analysis, placing observations into the phases of the innovation process and reading the result back against the literature.
The theory base is the startup innovation process as experimental development — Lean Startup and the build–measure–learn cycle (Ries 2011), customer development (Blank & Dorf 2012) — product–market fit and customer validation, and business model development, where the working tool is the Business Model Canvas (Osterwalder & Pigneur 2010). Cutting across all of it is the recent and still thin research on generative AI in entrepreneurship.
The study is looking for founders in the Oulu region who have used generative AI in developing their product and whose company is early-stage: growth-oriented, founded within roughly the last three years, still looking for a repeatable business model rather than scaling one. Teams still being formed, with a product in development, count.
Taking part is one interview of about an hour, in person or remote, at whatever time suits. There is no preparation to do and no material to send in advance.
Participation is voluntary and can be withdrawn. Recordings and identifiable transcripts are held only by the researcher, are not shared with the commissioning organisation, and are destroyed once the thesis has been accepted. Reporting is anonymised, and because the Oulu ecosystem is small, anonymisation is handled with more care than the usual formula — company names are pseudonymised at transcription rather than at write-up. Participants are given a written privacy notice under Articles 13 of the GDPR before the interview, and nothing is recorded before consent is given.
This is a Bachelor's thesis in Business Administration at Oulu University of Applied Sciences, supervised by the school. Founders of Oulu ry is in discussion as the commissioning organisation; if that is confirmed, the contact person will be a board member other than the researcher, since the researcher is an officer of the association. If it is not confirmed, the study runs as an independent research thesis and the results stay openly available to the region.
The finished thesis will be published in Theseus, the open repository of Finnish universities of applied sciences, and a summary will go to the founder community it came from. AI use in the research process itself is logged and reported openly in the thesis, which is both a requirement and, given the subject, the only defensible position.
Because the same question shows up in the client work. A company that buys software and gets its old operating model back, slightly faster, has made the same mistake as a founder who ships an AI-built product nobody asked for: the tool changed, the model did not. The study is that problem in its purest form, in the setting where the tooling moved fastest.
When there are results. The thesis will be published in Theseus under an open licence in spring 2027, and a summary will go to the Oulu founder community. Until then this page describes a question and a method, not an answer.
The method does, immediately — assumption registers, decision rules agreed before the data arrives, validation that is not a formality. The findings cannot, until they exist. If you see a claim on this site that founders do X, it will carry a citation you can check.
Connect on LinkedIn and say so. Interviews run from November 2026. You will get the privacy notice and a description of what happens to the recording before you decide, not after.
Read rather than cited from a reading list. The list will grow as the theory base is written.
If you are a founder in the Oulu region who has built with AI, the study would like an hour of your time. If you want the method applied to your own product rather than studied, that is the validation sprint.