This is a super classic failure mode and it happens more often than not, due to invisible failures in data pipelines. Ask me how I know ;) For example the connection to the actual database of… | Georg Zoeller | 11 comments

This is a super classic failure mode and it happens more often than not, due to invisible failures in data pipelines. Ask me how I know ;) For example the connection to the actual database of metric’s server could be silently failing and the model, optimized to please and deliver what the user requests just makes it up. Claude code users will recognize the pattern too, the AI just creating a placeholder test returning true or sitting an NPM module it fails to download to satisfy the user’s requests to make the test work on a system for which a library may not be available. The answer, from a CTO perspective, is to reinvest a significant part of the savings AI enables on code generation on superior observability rather than just allocating them to the shareholders. Observability is your only chance to meet compliance and perform your job as an executive or board member. Which is, as far as I can tell, not happening in most companies. You must demand it or a bunch of brain dead clawdbots will go gambling with your next earnings statement. | 11 comments on LinkedIn

linkedin.com