Home Newsletter

Don't Imitate Understand - #5

Hello!

In this fifth issue of the Don’t Imitate, Understand newsletter, I have a special announcement about a replacement for software development methodologies like Agile and Waterfall in the age of AI — and you can join what I hope to be a movement for software dev teams!

Agile and AI Don’t Work Well Together

LLMs have changed software development. They can generate features in minutes that once took days — and large amounts of code that’s subtly wrong.

Our development methodologies need to catch up. Typical methodologies like Waterfall or Agile don’t quite work in the AI age.

Waterfall’s rigid specs are too slow for AI’s rapid iteration. Agile’s minimal documentation provides poor context for LLMs and lacks concepts to mitigate AI risk.

Introducing Cascade Methodology

I’ve been developing software as a lead dev, architect, product lead, and UX lead for 25 years in a particular way. As AI-assisted development came into real use, I found that — unlike Agile and Waterfall — the way I’ve helped teams build software actually gets better with LLM assistance.

So, I am officially and freely releasing a new software development methodology called Cascade Methodology.

Cascades are rapid AI-driven development cycles that flow from detailed specifications through interactive prototypes to production code, with verification gates that scale based on business risk (entropy) tolerance.

It includes concepts such as:

• Minimum Solved Problems (MSP) — a problem-focused replacement for MVP.
• LLM-generated prototypes and user research as first-class citizens.
• Context awareness — understanding that AI requires good input context but its output can never be fully trusted.

To adopt Cascade Methodology means embracing and understanding both the benefits and dangers of AI-assisted development — and using them to your greatest advantage.

The 7 Core Principles of Cascade Methodology

1. Everyone has a seat at the table in writing specs.
2. Understand the underlying user and business problems the software should solve.
3. Build the least amount of software possible to solve those problems.
4. Prototyping and user research are first-class citizens.
5. AI-generated code is not to be trusted.
6. Code quality checks scale with the entropy tolerance of the business process.
7. Scope and timelines depend on entropy tolerance and problem understanding.

Cascades give LLMs the context they need while protecting against the risks of probabilistic code generation. It’s about embracing AI’s speed while staying grounded in solving real problems with quality software.

How Do Cascades Differ from Agile, Scrum, or Lean?

Ceremonies

• Agile/Scrum: Sprint planning, standups, grooming, retrospectives — often time-wasting and burnout-inducing.
• Cascade: Continuous spec writing and user research with everyone contributing. Specs become strong LLM context.

Estimates

• Agile/Scrum: Story points and forecasts — often guesswork and morale killers when missed.
• Cascade: Effort buckets based on entropy tolerance — the lower the tolerance, the more checks required.

Velocity

• Agile/Scrum: Story points completed per sprint.
• Cascade: Problems solved across a time period.

Measure of Progress

• Agile: “Working software is the primary measure of progress.”
• Cascade: Solved problems are the measure of progress.

LLM Code Generation Risk Mitigation

• Agile/Scrum: None — not designed for AI-assisted development.
• Cascade: Core concept — scalable verification gates for AI-generated code.

MVP vs. MSP

• Agile/Lean: MVP (Minimum Viable Product) — smallest product possible, often based on someone’s favorite feature.
• Cascade: MSP (Minimum Solved Problem) — smallest meaningful problem solved, which may not require new software at all.

Sometimes solving the problem doesn’t require more software — or requires far less than requested. Your software development practices should excite, encourage, and enable, not exhaust.

How You Can Learn More and Get Involved

The official GitHub repo for Cascade Methodology: 👉 https://github.com/AnthonyPAlicea/cascade-methodology

It includes links to my related blog posts, such as a long-form blog post on Cascades. I’ll also be writing a free-to-read book in that repo.

You can join the conversation in the Discussions tab — and if you like what you read, a star on the repo is greatly appreciated!

I’ll be producing YouTube videos and livestreams on this topic soon.

Can My Team Get Help With Cascades?

The methodology is free and open, but I also give talks and direct training to teams. One of the great things about Cascades is that you can adopt it gradually — one piece at a time.

Teams can start by improving:

• Spec writing for LLM context
• Integrating user research and rapid prototyping
• Measuring AI risk
• Shifting culture toward AI-assisted development

If you’d like a live Zoom consultation or workshop on any aspect of AI-assisted development and Cascades, just reply to this email with what you’re looking for.

AI-Assisted Software Development Is Now

Our existing methodologies weren’t designed for AI-assisted development. I invite you to read up on Cascade Methodology. Have questions? Drop them in the repo’s Discussions tab.