The H·AI·K·U Methodology

A universal framework for structured human-AI collaboration. Four phases that turn unstructured AI interactions into disciplined, repeatable workflows.

Why Most AI Collaboration Fails

Without structure, AI-assisted work suffers from predictable failure modes. H·AI·K·U addresses each one.

No persistent structure

Context lost between sessions. Every conversation starts from scratch.

No quality enforcement

Errors propagate unchecked. AI output accepted without verification.

No completion criteria

"Good enough" without verification. No way to know when work is actually done.

No mode selection

Wrong level of autonomy for the work. Too much oversight or too little.

No learning loop

Same mistakes recur. Teams never compound their experience.

No domain awareness

One-size-fits-all workflows. Security teams forced into dev sprints. Designers shoehorned into ticket queues.

How Stages Work

Within each phase, work flows through stages. Each stage has a specific structure: plan, build, review, and advance.

The Execution Loop

1
Plan

Agent articulates its approach before starting work.

2
Build

Agent executes the planned work, producing deliverables.

3
Review

Adversarial review evaluates work against completion criteria.

4
Advance

Quality gate decides: advance to next stage, or iterate.

Hats — Focused Roles

The AI agent switches between distinct roles during execution. Each hat constrains the agent's focus and behavior.

Planner

Reads requirements, plans approach, identifies risks.

Builder

Executes the plan, produces deliverables, runs verification.

Reviewer

Adversarial review against criteria. Cannot be the same agent that built.

Studios — Lifecycle Templates

Studios customize the lifecycle for specific domains. The default “ideation” studio works for everything. Specialized studios add domain-specific stages and review modes.

Collaboration Modes

Not every task needs the same level of human involvement. H·AI·K·U defines three modes — like a GPS that can be set to guide, monitor, or autopilot.

👁

Supervised

Human directs, AI assists

Best for high-risk work, unfamiliar domains, or critical decisions. The human drives; the AI is a thinking partner.

👀

Observed

AI executes, human monitors

Best for well-defined work with moderate risk. The AI does the work; the human reviews at checkpoints.

🤖

Autonomous

AI executes within boundaries

Best for routine, low-risk work with clear criteria. The AI works independently within defined constraints.