Large Language Models and Early-Stage Founders: An Exploratory Study Through the Lens of the Scientific Approach

Details

Supervisors

De Jaegere, Tanguy

Faculty

Louvain School of Management

Degree label

Master [120] en sciences de gestion, à finalité spécialisée

Abstract

enThis study examines how large language models (LLMs) are mobilized by early-stage founders, in relation to the four-step scientific approach to entrepreneurial decision-making validated by Camuffo et al. (2019, 2024). Rather than testing whether these tools improve venture outcomes, it asks how founders' actual LLM use aligns with, departs from, or reshapes the discipline this framework prescribes. Five semi-structured interviews with early-stage founders in Belgium and France were coded against Camuffo et al.'s (2019) four stages and five corresponding propositions (P1 through P5). LLM use was heavy at the earliest stage of venture building and around a set of risks cutting across all four stages, but comparatively rare at later, more structured stages. Four theoretically established risks, algorithm aversion, automation bias, cognitive offloading, and hallucination, were all present in the sample. A further pattern emerged: models tend to align their answers with what the founder appears to want to hear, rather than exercising independent judgment. Founders also differentiated tools by task and positioned the LLM very differently within their venture, from a supervised intern to a near-autonomous team member. This dissertation contributes to Camuffo et al.'s research program by examining general-purpose tools founders already use without training, and responds to calls for research grounded in a validated decision-making framework rather than a generic technology-acceptance lens.