That procurement framing is a useful filter, and the Gartner number about 40% of agentic projects getting cut by end of 2027 is going to age into a lot of post-mortems. Half the agents I see could be a scheduled function with one model call, and the agent framing is what makes them fragile and heavier to run than they need to be. What is the first question you ask to tell real autonomy apart from plain automation?