What if Agile is obsolete?
What if the forcing function is specification driven AI development?

In October 1999 Kent Beck’s book Extreme Programming Explained is published; in February 2001 the agile manifesto was signed. Both were driven by the very real need to remove the latency between specification and working software.

Prior to this software was developed using a variety of Waterfall methods - in simple terms comprehensive written specifications were created for a complete system, these specifications were then turned over to developers who created code to make the specification working software. The problem was the latency between specification creation and working software. The end users were required to imagine what they’d like the software to do and how they wanted the software to do it (what we would now call UX). The developers would then program the system and the end users would validate that the system did what the specification called for. But… Systems became increasingly complex and development time increased with that complexity. The latency introduced by this created a fundamental problem - it took too long for end users to realize the specification wasn’t actually what they needed or, increasingly, that the specification was no longer relevant to the work being done by the time the software was ready for validation. The solution was to decompose the specification into small, rapidly implementable units to remove the lag, generate rapid feedback and iterate toward the right specification and software system at the same time.

The forcing function that led to agile methodologies was the latency between specification and working software. The more complex the system, the more complex the specification, the longer the latency.

Fast forward to 2025.

AI-assisted development tools have fundamentally changed what’s possible in software creation. Large language models can now generate complete application components from detailed specifications, with AI agents capable of interpreting requirements, writing code, debugging errors, and even refactoring systems—all at speeds that would have seemed impossible even two years ago. Tools like Claude, Cursor, and GitHub Copilot aren’t just autocompleting lines of code; they’re building entire features from natural language descriptions. More significantly, a new pattern is emerging: Specification Driven AI Development, where developers create comprehensive specifications—capturing both intent (what to build) and blueprint (how to build it)—and then collaborate with AI agents to rapidly generate working systems.

What if agile’s original rationale is no longer valid? What if complex systems with complex specifications can now be turned into working software in weeks, not months or years? What if the quality of the specification - features, process, and non-functional requirements - is the driver of quality software? What if cycle time iterating on the complex system is days, not weeks?

Leveraging AI Agents and Specification Driven AI Development (with humans in the loop) can create complex systems with a velocity that was previously unimaginable removing the latency between detailed specification creation and validation. This same model can then also reduce iteration cycle time when a specification change is needed. In other words, if a comprehensive specification proves wrong, and it often will, the cost of that error drops dramatically when you can validate and revise in days rather than quarters. The imagination gap doesn’t disappear; it just becomes cheap enough to iterate through. This removes both constraints that drove Agile’s emergence:

  • Specifications no longer being accurate to what is needed due to system creation time
  • Specifications which are not actually what end users need (reality differs from imagination)

Assuming the early benefits of Specification Driven AI Development hold at scale, and the knock on effects (that’s another post) can be managed - is this the end of agile methodologies? Or will agile methodologies need to be completely re-thought and new modalities created to capture benefits further up the value chain?

About the Author: Brian Roy

I am a dynamic, straight-talking computer scientist, leader and entrepreneur dedicated to creating innovative products leveraging technology to solve real, meaningful problems. The views expressed here are my own and do not reflect those of my employers past, present, or future.

Author's Site