In a Medium blog post and accompanying LinkedIn post, Steve Jones, Executive VP at Capgemini, sparked intense debate by declaring that AI has killed the Agile Manifesto. Jones argues that agentic Software Development Lifecycle (SDLC) systems, where AI agents perform significant development work, fundamentally contradict the Manifesto’s four core values and twelve principles.
Jones identifies several critical challenges with applying Agile to agentic SDLCs. First, he argues that tools now matter immensely:
If you’re using Replit then its going to be different than if using Claude Code, and if you’re going to use a mix of Agentic SDLCs then you absolutely need to think about that.
This directly conflicts with the Manifesto’s preference for individuals and interactions over processes and tools.
The speed differential presents another core issue. Jones describes creating working applications in hours and migrating entire applications on single flights, stating:
This is where Agentic SDLCs are causing a fundamental break with the Agile principles, because Agentic SDLCs are too fast for Agile.
The traditional two-week sprint cycle appears antiquated when AI can generate functional code in minutes.
Perhaps most significantly, Jones challenges the principle of working software over comprehensive documentation. AI agents excel at creating software that appears functional but can accumulate technical debt at unprecedented rates. He notes that AI is highly effective at building software that appears to work, or at least works for the very specific instructions it was given, making documentation and architectural planning more critical than ever.
The reaction has been mixed. Rolf Läderach, a Head of Operational Excellence & Agile Coach at Sandvik, countered in the LinkedIn discussion that:
Agile is not the Manifesto, and it is certainly not about frameworks. Agile is about creating adaptive and learning organisations that can respond to change and deliver outcomes. That need will never disappear. And AI supports this.
Sonya Siderova, CEO of Nave, offered a nuanced perspective:
Agile isn’t dead. It’s optimizing a constraint that moved.
She argues that while standups coordinate human work and retros extract human learning, when agents handle building in minutes, the bottleneck shifts from “how do humans collaborate to build” to “how do humans decide what to build and validate it actually works.”
Kent Beck, one of the original Agile Manifesto signatories, has been exploring what he calls “augmented coding” as distinct from “vibe coding.” In a detailed Substack post, Beck describes augmented coding as maintaining traditional software engineering values—clean code, comprehensive testing, and careful design, while allowing AI to handle much of the typing. He distinguishes this from vibe coding, where developers simply feed errors back to AI, hoping for fixes without caring about code quality.
Beck’s approach suggests a middle path: using AI as a powerful assistant while maintaining engineering rigor. He reports building a production-ready B+ Tree library in Rust and Python using this methodology, with AI generating code under careful human supervision guided by Test-Driven Development principles.
The industry response extends beyond individual commentary. Casey West has proposed an Agentic Manifesto that adapts the original Agile values for autonomous AI systems, shifting from “verification” (did it do what I said?) to “validation” (did it do what I wanted?). Multiple organizations are experimenting with “Agentic Delivery Lifecycles” (ADLC) that wrap traditional SDLC practices with new governance models for non-deterministic AI behavior.
AWS has echoed this in its 2026 prescriptive guidance, suggesting that “Sprint Planning” must evolve into “Intent Design.” In this model, architecture becomes “scaffolding”, defining roles, guardrails, and fallback mechanisms rather than scripting every decision path.
However, Forrester’s 2025 State of Agile Development report presents a striking counterpoint: 95% of professionals affirm Agile’s critical relevance to their operations, with 61% reporting deployment of agile practices for over five years. Diego Lo Giudice, VP and Principal Analyst at Forrester, notes in a Forrester blog post that while proficiency levels vary, only 7% achieve full proficiency, the integration of agile with generative AI offers a promising avenue for enhancing agile’s impact even further, with nearly half of respondents already leveraging genAI in their agile practices.
Jones himself acknowledges he’s “not saying every ‘agile’ practice now has no value,” but insists the methods designed for weeks-long development cycles by human teams don’t translate to agent-driven development. He calls for a new manifesto and methods based on the assumption that agents will handle much of what developers previously did.
The debate raises fundamental questions: Is Agile a specific methodology tied to particular practices like sprints and standups, or a broader philosophy about adaptability and learning? When AI can generate code in minutes, what becomes the true bottleneck in software development? And do we need entirely new frameworks, or can existing Agile principles evolve to govern human-AI collaboration?
As Eric Newcomer, an Analyst, Advisor, and Security architect, commented in the discussion:
I don’t know, I can agree we need a new manifesto all right but I think bureaucracy killed agile before AI agents came along.
What seems clear is that software development is entering a period of significant methodological upheaval. Whether this represents Agile’s death or its evolution into something new remains an open, and increasingly urgent question for the industry.
