Observation
Agentic AI can perform exploratory data anslysis (EDA) and data cleaning within a few minutes. Questions can be asked and answered by the analyst without directly ever inspecting the dataset.
Tension
EDA was never just a step, it was how human intuition about the dataset was built. Now the process is partially invisible, and results are accepted before the data is understood.
Shift
This aligns with the scientific method as defined by Deutsch/Popper: form a conjecture, test it against data, refine the explanation, and generate new questions.
- Analytics is no longer: clean → explore → analyse (and done.)
- It becomes: ask → accept an approximate answer → improve the data pipeline → repeat
_Understanding emerges through iteration, not inspection._
Implication
The dataset is no longer the product. The pipeline is.
Working belief
1. Data will never be clean or complete. It is continuously approximated and refined. 2. Agentic analytics works if iteration and validation co-evolve.
Next step
Add explicit feedback loops into ProcurePlus (Menacon) including:
- thumbs up/down on matches on ProcurePlus (Menacon)
- ever improving glossary for the tender scraper, born out of tender analysis