Coffee & Code: Just shipped a feature that automatically improves the data context layer, as I posted about yesterday.
Worth writing about this a bit more. Everyone thinks data documentation and business logic is super important for AI agents, and it is! Yet, at many (most?) orgs this stuff lies in ruins and excavated now and then by diligent analysts.
We need a better and systematic way to address this gap than to wish for behavior and cultural change. Hence, the blueprint laid out in this feature. Copy it if you are building agentic knowledge-work systems. This kind of auto-improvement loop is incredibly effective.
Auto-context improvement is now live in VerbaGPT!
The way it works is that you simply query a connected database. The custom harness may stumble and likely recover. But the stumbles get logged and a daily job runs in the background that looks at where LLM failed to get it right the first time, and see if the context around the database (data dictionary, business logic etc.) can be improved. It will then surface these improvements for you as the user to review and approve.
All done with correct security and permissions.
Have not tried VerbaGPT yet? There are several demo databases available on VerbaGPT Online, covering machine learning, healthcare data, weather data, and SEC 13F (stocks) data.