Should we revisit Extreme Programming in the age of AI? Author: Jacob Clark, Engineering Director Date: 01 Sept 2025 Reading time: 9 mins --- Overview The software development landscape has drastically accelerated with AI and modern platforms enabling rapid code generation. However, faster output has not translated into better delivery outcomes: many projects still fail to meet expectations, budgets overrun, and users remain underserved. This suggests that the bottleneck is not code production speed but something else. --- Key Points Output is Not the Problem Over decades, software development has become faster via: High-level languages Frameworks and package managers DevOps and serverless computing Developer platforms abstracting infrastructure AI-enabled code generation Despite these innovations, studies (e.g., Standish Chaos Report and McKinsey) show persistent failure and underdelivery in IT projects. The real need is not more acceleration but smarter constraints to focus on learning, alignment, and purposeful building. XP as a Counterweight Extreme Programming (XP), from the late 1990s, is designed to introduce deliberate friction and constraints, facilitating learning over raw throughput. Practices like pair programming reduce output in the short term but increase shared understanding, trust, and quality. XP emphasizes going slower in small increments to achieve faster, better long-term outcomes. XP shapes both collaboration and code; it focuses on direction and intent rather than just speed. AI Magnifies Problems XP Was Built to Solve AI can produce code faster than it can be validated, risking layers of unvetted or fragile logic. Autonomous AI agents may compound complexity and assumptions without human oversight. Research shows large language model (LLM) accuracy degrades over longer context windows, resulting in brittle code over time. XP was created to prevent these kinds of runaway entropy and maintain software quality. Software is Still Human Software remains a human-centric discipline affected by culture, incentives, communication. Barriers such as alignment, shared context, clarity, and user validation persist regardless of tooling. XP values remain relevant: Simplicity: reduce complexity Communication: maintain team cohesion Feedback: enable learning and adaptation Respect: build trust and safety Courage: allow transparency and change From Feature Factories to Value Delivery Successful teams prioritize flow and feedback over raw velocity and features. XP practices like small batches, continuous integration, automated testing, and shared ownership help teams stay adaptable, resilient, and user-focused. These practices will be critical as AI accelerates code output, ensuring quality and risk management. Lessons from the Past Chaos Report project success rates: 1994: 16% delivered on time and budget 2012: 37% success 2020: 31% success (regression) Despite innovations (Agile, DevOps, cloud, AI), delivery reliability has improved marginally. Toolchains alone are insufficient—methodology matters. What Needs to Change? Output creation is no longer the bottleneck. Faster code is meaningless without validation and alignment to real needs. Invest in outcome-driven capabilities: stronger feedback loops, clear product direction, tighter collaboration, design discipline. Make the process more human-centered. Even with AI, sustainable delivery depends on human collaboration. Hyperact’s Product Operating Model aligns product strategy, operating rhythms, and engineering practices around people for better delivery. --- Conclusion: Should We Revisit XP? Yes. XP anchors software development in human-centered practices amid rapid tool acceleration. It emphasizes discipline, empathy, team