- Author
- Jacob Clark, Engineering Director at Hyperact
- Published
- 1 September 2025
- Kind
- Consultancy blog post, argument with cited statistics
- Original
- www.hyperact.co.uk/blog/should-we-revisit-xp-in-the-age-of-ai
- Subject
- Coding with agents
What it is
Jacob Clark is an Engineering Director at Hyperact, a UK product consultancy. The post asks whether XP helps teams that now generate code with AI. XP was developed in the late 1990s.
What it says
- The post says output is no bottleneck: code is cheap but delivery stays poor.
- It says Extreme Programming adds deliberate friction, such as pair programming, so teams learn.
- It claims AI magnifies the risk of producing software faster than it can be validated.
- It claims LLM accuracy declines as context grows, so heavy vibe coding yields brittle code.
- It cites CHAOS on-time, on-budget rates: 16% in 1994, 37% in 2012, 31% in 2020.
- The post concludes the process must become more human and answers yes to revisiting XP.
What evidence it gives
The evidence is argument plus cited statistics. These are the Standish CHAOS study and a McKinsey figure of 70% failed transformations. It also cites a research claim on context windows. It shows no data from teams using XP with AI.
Interests and limits
Hyperact sells product consulting and the post promotes its Product Operating Model. The CHAOS and context-window claims are cited without links followed here. The post does not test whether XP practices improve outcomes with AI coding tools.