In programming problem-solving, learners engage in creating solutions to specific given task problems by writing an exe-cutable code. Prior research has shown that self-regulated learning (SRL) strategies help improve novice performance in solving programming problems. With the advent of Large Lan-guage Model (LLM) tools like ChatGPT, novices can generate reasonably accurate code by providing the problem prompt. They, hence, may forego applying essential self-regulation strategies such as planning and reflection to solve the problem. This research investigates if the above is the case. We designed a programming problem-solving task in an available online environment, LAreflecT. A set of self-regulation prompts was provided while learners could use ChatGPT to build their so-lutions. Learners¿½f interactions with the elements in the LAre-flecT and their generated artefacts are logged. We analyzed 42 undergraduate students' data and highlighted problem-solving approaches of groups with correct and incorrect end-point solutions through process mining and artefact analysis. The findings indicate when learners use LLM as support their epis-temic actions involve refining their problem understanding and solution evaluation when supported with metacognitive prompts within the system. We discuss the reflections of the learners who had more than two conversations with ChatGPT and draw implications of designing SRL support while learn-ing with generative AI.