<p>This study used an instrumental approach (Drijvers et al., 2013) to investigate how artificial intelligence (AI) or large language models (LLMs), specifically ChatGPT 3.5, transitioned from an artifact into an instrument for four prospective secondary mathematics teachers through the development of schemes and techniques in using the AI to construct teaching materials, a lesson plan, student activity, and homework assignment for a lesson on solving quadratic equations. In using a LLM, the user’s observable actions, or techniques from an instrumental approach consist of the prompts entered into the chat window. The participants identified their own criteria for high-quality teaching materials before prompting the AI to generate lesson materials aligned with these criteria. The number of techniques the prospective teachers used ranged from as few as three to as many as twelve. The prospective teachers’ justifications for their revised techniques led to a set of conjectured instrumentalization and instrumentation schemes. Three instrumentalization schemes appeared among the four prospective teachers: organization, specificity, and generalized LLM. Five instrumentation schemes appeared among the four prospective teachers: shifting content focus based on AI limitations, organization, dispensing with specialized vocabulary, specificity, and settling. The results showcase techniques and schemes that prospective secondary mathematics teachers demonstrate when using AI in lesson development and have implications for supporting teachers in productive AI integration.</p>