Waterfall is back
For decades, the battle between Waterfall and Agile methodologies defined how teams delivered projects—one rigid and predictable, the other flexible and adaptive. But now, with AI-powered tools turning hours of coding into minutes, even Agile might not be fast enough.
The Shift from Waterfall to Agile
Software development began with Waterfall—a linear, phase-based approach (Requirements → Design → Implementation) suited for predictable projects. But as change became constant, its rigidity faltered.
Agile emerged as the answer, favoring iterative cycles, adaptability, and continuous feedback over exhaustive upfront planning. It thrives on uncertainty, delivering value incrementally while welcoming evolving needs. Yet, this flexibility shouldn’t excuse poor design—good software, like timeless architecture, balances simplicity with forethought.
Ironically, Waterfall resembles a single, drawn-out Agile iteration, with MVP be the full and final product.
LLM + Agents: A Game Changer?
Recently, I experimented with LLM-powered development tools for hobby projects. The experience was exhilarating—I could rapidly prototype ideas, pivot when needed, and embrace Agile’s essence. But a glaring issue emerged: without proper documentation or design, every adjustment felt like starting from scratch. Instructing an LLM to modify an unfamiliar (and largely auto-generated) codebase proved frustrating.
After a week-long break, I realized the missing piece: structured planning. I revisited traditional practices—writing an SRS (Software Requirements Specification), drafting a System Architecture document, and outlining a Technical Design. Surprisingly, LLMs excelled at expanding on my initial inputs, generating coherent documentation and even breaking tasks into a Work Breakdown Structure (WBS). By committing progress incrementally in Git, I avoided the "vibe coding" trap and regained control.
This Time, It’s Different
Why return to "old-school" documentation? Because AI accelerates iteration to an insane degree. What once took a sprint (two weeks) now takes minutes. Agile’s feedback loops—designed for human-paced collaboration—break down when you could, in theory, "complete" multiple sprints in an hour.
If we stick to bi-weekly cycles but deliver far more features per iteration, we introduce greater uncertainty and assumptions. To mitigate this, I can only think of two options:
- Hyper-fast sprints (e.g., daily iterations. like Charlie Chaplin in Modern Times? )
- Slower, deliberate planning—embracing Waterfall’s thoroughness while ensuring designs accommodate uncertainty.
AI-Era : Where Rigor Meets Speed
AI supercharges Agile's speed, but demands Waterfall's rigor:
- Sharper Requirements – Clear specs prevent AI from magnifying chaos
- Thorough and Adaptive Design – Build modular systems ready for inevitable changes
- Structured Tracking – Version control and docs act as AI's "undo button"
Agile's adaptability remains crucial, but AI requires us to build thoughtfully at speed – creating software that's both rapid and resilient.