NN/g: agentic AI systems are turning into balls of mud
Cheap AI building lets people grow agent setups piece by piece until nobody can explain them.
Why it mattersReported: the article argues that when building is nearly free there is no scarcity forcing planning, so systems expand through patches and throwaway code, and users come to depend on the AI to navigate their own setup. It recommends keeping global, local and ambient context separate and labeled, auditing global context regularly, and rebuilding once you can no longer say what each part does. It reads as an argument piece; no measured data is cited. Opinion: scarcity was an unnamed constraint that did useful work, and removing it recreates legacy complexity at speed.
Go deeper