This is H·AI·K·U (formerly AI-DLC) — a methodology for structured, iterative software development with AI assistants.

The challenge

Working with Claude changed how we write software. It also broke things we used to take for granted.

WHAT BROKE

Context shifts mid-task. The agent loses track of which problem it's solving while it's solving it.
Output volume outruns review. The model writes faster than any human can read carefully.
The old metrics stop saying much. "How's it going" turns into a guess.

We built H·AI·K·U because the answer to all three couldn't be "trust the model."

The solution

H·AI·K·U introduces two core concepts: hats and units.

Hats

Hats represent distinct mindsets and responsibilities. By explicitly switching between them, you prevent context drift and make sure each phase gets the attention it needs.

Researcher

Understand the problem before acting. Read the code, ask the questions, surface the constraints.

Planner

Design the approach with clear steps. Decide what done looks like before anything gets built.

Builder

Execute the plan with focus. No drift into adjacent problems, no scope creep.

Reviewer

Validate quality and completeness. Catch what the builder missed before it ships.

Units

Units are focused pieces of work with clear success criteria. Breaking work into units does four things at once.

Measurable progress

Each unit has a clear definition of done. You know what shipped and what didn't.

Controlled scope

The unit's criteria define the edges. Drift into adjacent work is visible immediately.

Verifiable quality

Gates run per unit, not per intent. Bad work gets caught close to where it was made.

Sustained momentum

Small wins land while big work is still in flight. The flywheel never stalls.

Getting started

H·AI·K·U is distributed as a Claude plugin. Install it with:

/plugin marketplace add gigsmart/haiku-method
/plugin install haiku@gigsmart-haiku-method --scope project

Then start using the hat commands to structure your workflow.

A Claude plugin

H·AI·K·U is a Claude plugin that works in both Claude Code and Cowork.

What's next

What's on the bench: more docs and guides, CI/CD integration, team collaboration features, and the metrics surface that finally tells you whether a run is going well.

Follow along on GitHub.