Completeness in AI Results
retained text
I’ve been working to develop a number of agents that perform different planning types of activities and evaluations. One thing that I’ve noticed is that there seems to be not only an issue with accuracy in AI (aka Hallucinations) but also in completeness of answers sometimes. That is to say either cutting off results before they are finished or conclusive or skipping steps or elements in complex tasks.. The interesting thing though is that it seems like once you make the AI aware of the missing details, steps or information that it does immediately recognize it as valid. Similar to when you catch it in a hallucination and it responds. “Oh yeah, you’re right. I’ll fix that.” I was curious to ask of this group. What do you do in your agents to ensure completeness in results? Not accuracy, though that is important. But completeness. My current methodology has been a mixture of looping using different models within a given platform or comparing results through multiple platforms and iterating on the response before it’s given. For some specific examples of this, if I give you a complex project, could you define all of the steps in between from beginning to end to say that that project has been completed? Then could you further break that project work down into its sub elements. What I found in the results is that it can miss elements of the work breakdown, structure, or factors for consider consideration in the requirements of completion. This would be like if you asked it to give you a project plan for building a house, and it left out the electrical and plumbing. You know that it knows that electrical and plumbing are considerations for a house but trying to figure out why did it not take that into Account originally and how did it miss that? My career has been as a project manager and data analyst and so yes, I am primarily referring to a Work breakdown structure and requirement gathering, or definition of done. The results of AI sometimes seems to be similar to that meme or short about the kid describing for his father, the steps to make a sandwich. I don’t want to be that kid, frustrated at getting the wrong results when perhaps my process or input should have been made clearer.
document fetched 2026-10-13 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x4176t/completeness_in_ai_results