ai-agile-talk

Last updated: January 23, 2026 at 12:51 AM

ai-agile-talk


About: Et's talk on how AI enables a return to waterfall-style development through spec-driven development
Started: January 23, 2026


12:19 AM


Core Thesis: AI Enables Return to Waterfall


The Agile Problem Agile Solved

  • Humans are bad at complete specifications upfront
  • Requirements change, assumptions are wrong
  • "Fail fast, iterate" was the solution to human cognitive limits

    • How AI Changes This

  • Spec-Driven Development: AI can help create comprehensive specs
  • AI can think through edge cases humans miss
  • AI can validate requirements and catch gaps upfront
  • One-shot implementation becomes viable again

    • The Argument

    Agile wasn't inherently better - it was a workaround for human limitations in specification and planning. AI removes those limitations.

    12:22 AM


    The Code Generation Trap


    Founder Misconception: Code generation = simplicity Reality: More code = more complexity by definition

    The Problem

  • Founders see Claude generating thousands of lines instantly and think "this is simple"
  • They confuse *generation speed* with *solution elegance*
  • AI tools can generate complex, bloated solutions very quickly
  • Just because it's easy to generate doesn't mean it's good code

    • The Danger

  • Code generation without constraints leads to over-engineering
  • AI will happily build a 50-file architecture for a 5-file problem
  • Volume of code != quality of solution
  • More code = more bugs, more maintenance, more technical debt

    • The Counter-Argument

    This is why spec-driven development is crucial - the spec needs to emphasize *constraints* and *simplicity*, not just features. AI needs guardrails to generate elegant solutions, not just working ones.

    ai-kills-agile-talk


    About: Talk prep: "Agile is Dead Because Waterfall Killed It Because of AI" - exploring how AI enables spec-driven development and makes upfront design viable again
    Started: January 23, 2026


    12:20 AM


    Core Thesis

    "Agile is Dead Because Waterfall Killed It Because of AI"

    The Argument

  • Agile wasn't inherently better methodology - it was damage control for human cognitive limitations
  • Humans are bad at specs: miss requirements, wrong assumptions, can't think through edge cases
  • "Fail fast, iterate" was because we couldn't get it right upfront
  • AI changes this: can help think through complete specs, handle "what if" scenarios, catch gaps
  • Spec-driven development becomes viable again with AI as co-pilot

    • AI as Constitutional Document

  • Spec keeps AI grounded in initial thinking
  • Prevents context rot and hallucinations
  • Without solid spec, LLM wanders into "Frankenstein monster" territory by iteration 10
  • Strong upfront spec keeps AI "on rails"

    • Controversial Angle

  • Challenges 20+ years of Agile evangelism
  • People have built careers on Agile methodology
  • Suggests we're going back to something that "failed" before

    • 12:51 AM


    System Prompts: The Invisible Foundation


    Key insight from building AI orchestration system: Models in their "basic state" are almost useless. The system prompt is everything.
    Founder Disillusionment:
  • Think models "know everything" out of the box
  • Reality: Raw GPT-4 without proper prompting is like hiring a brilliant person with amnesia
  • The magic isn't in the model - it's in the engineering around it

    • System Prompt as Critical Infrastructure:
  • Defines personality, working style, constraints
  • Sets context and domain knowledge
  • Establishes quality standards and output format
  • Without it: generic, unfocused, often unhelpful responses

    • Connection to Waterfall Thesis:
  • Just like specs ground AI code generation
  • System prompts ground AI behavior and decision-making
  • Both require upfront investment in definition and constraints
  • Both prevent AI from wandering into irrelevance

    • The Deception:
  • Founders see Claude/ChatGPT demos that work well
  • Don't realize those demos have carefully crafted prompts
  • Think they can just point raw AI at their problems
  • End up disappointed when AI gives generic business advice instead of domain-specific solutions

    • ← Back to all brainstorms