After the whole euphoria of using Agentic AI as shown at #JAX2026 in Mainz, I've tested in my project where I am hitting a long time issue with HAPI FHIR.
Result: after more than 6 hours running and trying to find the issue, it ended up cheating by modifying the output error, downgrading it from error to info, so no validation issue happens.
I've ended it by reverting the code to last commit
My personal 2cents, you can vibe all you want, but give AI a legacy project, with more complicated frameworks behind that and you get a receipt for chaos, frustration and loss of money.
The whole health industry is based on FHIR (or plans to move to it), but the reality is that only a few individuals are maintaining the core libraries, making opinionated contributions and sometimes changes aren't properly clear
Lenny Primak
Michele Adduci