In this context, making a broad claim such as 'ChatGPT is bad for learning' (e.g. Kosmyna et al. 2025) makes 1. Atwo-load model: It returns to a two-load model, being used for and in what context. A technology like construct. The effortful process of schema Al is entangled with learning and thinking in numerous, complex ways, driven by many factors, not least the load. Thus, there are only two loads: intrinsic learner's goals and motivations. Incorporating some (unavoidable) and extraneous (non-essential). 2. A“goal-driven" approach: This is the most with sound and poor use of Al in learning and teaching. critical modification. Kalyuga and Plass where the learner's activities, motivations, and goals, not just the instructional design, individual differences, particularly related to goals. In a study of 1,445 Al-assisted writing sessions, learners understanding the challenges Al poses to learning. in their work. Conversely, learners who passively different goals. Learners, especially novices, are often saw a decrease in quality (Yang et al. 2025). The risk of motivated by a desire for efficiency (Zhai et al. 2024), which translates to a task completion goal. Other incentivise the task completion goal, leading learners, or the same learner in a different context, students to frame the intrinsic load of learning, the may have a ‘learn the concept' goal. The learner's goal effortful thinking, as something to be outsourced, determines how they interact with the technology. rather than embraced. This is particularly the case cases for Al in real learning settings. As has long to be a negative experience (David et al. 2024). been argued by Fawns (2022), Dron (2022), and others, technology use in education cannot be represented by a singular, monolithic use case.