Anti-Patterns

This chapter catalogues the ways in which enterprise AI programmes fail. Chapter 18 showed five composite cases that worked, at least eventually, and Chapter 19 one real incident in which several controls failed at once; this chapter is about the far larger population of programmes that fail more quietly. Each anti-pattern is described through its symptom and cause, its consequence and its remedy, so that you can recognise it early, name it without causing offence, and point to the chapter that holds the corrective practice. The catalogue extends the anti-patterns in Architecture as Code (Chapter 26B) (66, Nordqvist, 2025), many of which reappear here in AI-specific form.

An anti-pattern is not a one-off mistake but a solution that looks reasonable, is adopted repeatedly, and produces the same bad outcome every time. Sculley and colleagues (48, Sculley et al., 2015) described the first generation for machine learning as hidden technical debt such as entanglement and pipeline jungles. Paleyes, Urma and Lawrence (55, Paleyes et al., 2022) found the same deployment obstacles recurring across case studies. The anti-patterns here are their organisational and architectural descendants, grouped as portfolio, governance and engineering anti-patterns.

Several anti-patterns are mode errors: a system in one usage mode of Chapter 2 is governed as if it belonged to another.

Primary topics explored

1. How to read this catalogue
2. Portfolio and strategy anti-patterns
3. Governance and compliance anti-patterns
4. Engineering and operations anti-patterns
5. Detecting anti-patterns as code
6. Summary