Truthfully have been a little skeptical of how many workloads will actually need “agents” vs doing something totally deterministic with a little LLM augmentation. Seems like I’m not the only one that thinks the latter works a lot of the time!