A Towards Data Science article described an attempt to build an “agentic AI network” for advising manufacturing plants using uploaded assessment data.
The author wrote that early prototypes produced plausible but incorrect and fabricated numerical outputs.
The author argued deterministic execution is required for foundational analytics and proposed separating deterministic data analysis from LLM-based reasoning and interpretation.
The piece described an implementation in Microsoft Copilot Studio where a parent agent orchestrates sub-agents and an analytics module for reliable processing of complex Excel exports.
Source: Towards Data Science