AI agents grow incrementally into fragile systems

Summary

NN/g argues that the ability to set up AI and agent-based systems at near-zero cost has led non-engineers to assemble these systems piece by piece. This results in fragile 'balls of mud' that become unidentifiable and unrefactorable due to accumulated complexity.

Why it matters

The article identifies a structural failure mode in the current AI agent ecosystem: the elimination of historical scarcity constraints—time, expertise, and capital—that previously enforced architectural discipline. It warns that the era of 'vibe coding' is recreating decades of software engineering mistakes on a large scale. Those most affected are non-technical developers who lack the terminology to realize their systems have become unmanageable.

Highlights

• AI eliminates the scarcity constraints (time, money, expertise) that previously forced prioritization and architectural planning. Developers overbuild because it 'feels productive.'
• Non-engineers are building complex agent workflows guided by AI, X, and Reddit. This process lacks release cycles or a state of 'completion.'
• Observed failure patterns align with the 1997 Foote & Yoder 'Big Ball of Mud' classification. This demonstrates that AI cannot escape classic software engineering failure modes.

Critical take

The core claim is merely a repackaging of a 1997 observation with AI added as a new variable. The '5 patterns' framework carries the risk of post-hoc rationalization of unstructured development. More critically, while NN/g diagnoses the ailment, it offers no concrete remedies beyond 'plan your architecture.' This is a skill that the aforementioned 'vibe architects' cannot implement precisely because they have not received training in it.