For an AI coaching agent to genuinely support adults in their learning, it is not enough that the language model produces fluent text. The dissertation identifies where the real hurdles lie: anchoring such coaching agents in established coaching theory, preventing them from becoming unnecessarily complex and expensive, and ensuring that learners actually engage with them. Across seven papers, each of these hurdles is translated into concrete design decisions with measurable effects. What is evaluated is not the language model in isolation, but the coaching agent in use with learners.

Supervisor and first advisor Jan Marco Leimeister and second advisor Emanuel de Bellis offer their warm congratulations and ceremoniously present the doctoral hat.