This paper discusses the potential of MyGPT - a configurable bot developed with the GPT-4o model as an active instructional tool in physics education. Guided by the scaffolding strategy, we created the PhysicsBot with MyGPT tool and tested it in a simulated environment regarding its potential to achieve better learning outcomes. The general objectives of this research were threefold: exploring how effectively PhysicsBot could open and sustain dialogues, provide adaptive feedback, and foster autonomous problem-solving skills. These observations confirm that MyGPT works well to support collaborative learning and to engage in adaptive management of the level of the individual Zone of Proximal Development (ZPD) with customized guidance. However, a number of these challenges have emanated during the process related to smooth adaptation to the evolving ZPD and gradual reduction of guidance, hence suggesting areas for improvement. Higherorder bots that assess the ongoing discussion in a second, invisible to the user, thread can be developed using GPT-API. This allows for a thorough grading of conversations and refinement of the employed instructional methodology.