This trend towards heterogeneous computing, whereby CPUs, GPUs and FPGAs are working in concert, is not justaboutpower but about ensuring that valid systems can be built for these applications like AI training or data-intensive research. Yet, schedulingtasks to combine these heterogeneous resources is still a challenging problem. Static techniques such as HEFTaresusceptibletounexpected and varying conditions; dynamic heuristics like Min-Min generally focus to near optimal solution, but lacktrueadaptability.Studies (e.g., [47, 56]) have demonstrated that these kinds of applications may incur performance inefficiency as high as over 20%whenthe workloads modulate itself, which brings into question the robustness of existing approaches. Reinforcement Learning(RL), despiteitsflaws, also provides some hope in learning on-the-fly rather than rote-acceptance of stability. The average completion times for largescaleGPU clusters have reportedly dropped below 320 ms through DRL-based schedulers, as opposed to over 400 ms bytheHEFT.Thisdifference means less wasted cycles and lower cost per joule. However, RL models propose challenges: they are expensivetotrainandtend not to be transparent. This paper contends that self-adaptive AI scheduling is a direction full of promises but that needs tobetreadedcarefully and with the right degree of expectations. Keywords: Task Scheduling, HEFT, CPU, GPU, Field Programmable Gate Arrays (FPGA), Self-Adaptive, Reinforcement Learning,Heuristics, Deep Reinforcement Learning (DRL), Min-Min Scheduling, Round Robin, AI Scheduler, HeterogeneousComputingArchitecture (HCA), Cloud Computing