B. Prendergast

renderg.host

20 September 2026 14:34

I'm passionately against AI slop. I also use LLMs constantly to make tools and workflows art. To me, there is no contradiction because 'using AI' isn't one activity. Writing a prompt and building tools are different acts, with a fundamentally different relationship to conceptualisation and creation.

Joe Casabona

casabona.org

01 September 2026 23:38

As well you should :-) For me the core is use AI for tasks and not thinking. I suspect one could argue that writing code straddles that line by coding harnesses are also very good. I do have a longer article here about it: streamlinedsolopreneur.com/ai-brain-rot/

Jeremy Nelson

jeremynelson.bsky.social

26 February 2026 03:51

Exactly. 'No ETL' just means 'proprietary ETL you can't see or fix.' I've watched teams trade portable SQL for vendor lock-in and then struggle when the 'magic' breaks on edge cases. The transformation logic always exists somewhere—the question is who controls it.

Jeremy Nelson

jeremynelson.bsky.social

24 February 2026 13:13

This is the truth that nobody wants to hear. I've spent days tuning prompts and model parameters only to realize the training data had a subtle encoding issue. Now my first debugging step is always 'check the data' before touching any model code.