HierarchyLooping

HierarchyLooping is a Markdown-first operating system for human-AI business workflows.

Many AI workflows start as one-off prompts. That often creates inconsistent quality, unclear responsibility, weak handoff between people and AI, and little reuse of what was learned from previous successes or failures.

HierarchyLooping solves this by turning work into a governed improvement cycle. It defines the purpose of the work, the input that starts it, the steps to follow, the success conditions, the constraints, the approval rules, the evidence to record, and the improvements to consider for future runs.

It is designed for broad business operations, not only software development. A workspace can represent customer support, meeting operations, finance review, recruiting, compliance, procurement, personal operations, or any other repeatable business activity.

HierarchyLooping organizes work into ten layers:

0. Signal / Event
   A real-world trigger, such as a customer request, meeting note, scheduled time, file change, email, or external system event.

1. Atomic Action
   A small individual action, such as reading a document, classifying a request, extracting facts, drafting a short response, or checking a status.

2. Task Loop
   A complete task made from multiple small actions, such as preparing a meeting brief, organizing a reimbursement request, or summarizing a support ticket.

3. Execution Loop
   A repeatable operational workflow, such as daily inquiry triage, weekly review preparation, invoice exception handling, or scheduled personal task review.

4. Domain / Workflow Loop
   A broader business workflow that coordinates multiple execution loops, such as customer operations, finance operations, hiring operations, compliance operations, or personal life administration.

5. Improvement Loop
   A review layer that studies repeated results, failures, delays, manual corrections, and approval patterns. This layer proposes improvements to workflow definitions, checklists, prompts, schedules, and review criteria.

6. Loop OS / Governance Loop
   A governance layer that controls permissions, risk boundaries, approval rules, stop conditions, automation levels, workspace compatibility, and release safety.

7. Organization Learning Loop
   A learning layer that examines how people and AI should divide responsibilities, what training or documentation is needed, and which practices should become shared organizational knowledge.

8. Strategy / Portfolio Loop
   A portfolio layer that decides which workflows should receive more investment, which should stay manual, which should be automated further, and which should be retired.

9. Meta-Strategy Loop
   A top-level layer that reviews the improvement philosophy itself: whether the current governance model, automation strategy, and evaluation approach are still appropriate.

Improvement happens through evidence moving upward and decisions moving downward.

A lower layer run does not rewrite the system by itself. It records what happened, what evidence supports the result, what worked, what failed, whether human approval was needed, and what improvement signal should be considered. Repeated issues, successful patterns, and risk observations are then reviewed by higher layers.

For example:

• If a single support ticket is unclear, the Task Loop may ask a human for clarification.
• If the same kind of unclear ticket appears many times, the Execution Loop may reveal a repeated pattern.
• The Improvement Loop may propose a better intake checklist or routing rule.
• The Governance Loop checks whether the change affects risk, approval, or automation level.
• A human or approved higher-level decision accepts, revises, or rejects the change.
• The updated workflow is versioned and used in future runs.

This means improvements are gradual and controlled. The system can start by only observing and drafting. After enough successful runs and reviews, it may move toward limited low-risk execution. Higher-risk actions still require human approval, and any increase in automation must be justified by recorded evidence.

Key capabilities:

• Define business workflows using readable Markdown files
• Separate the reusable engine from project-specific workspaces
• Organize work across layered loops from events to governance and strategy
• Route work through explicit run packets and canonical adapters
• Record results, evidence, human decisions, and improvement signals
• Keep execution, evaluation, and approval as separate responsibilities
• Support chat-based starts, scheduled starts, and external event triggers
• Review repeated failures and operational patterns over time
• Propose improvements based on evidence rather than one-off impressions
• Control automation level through risk, maturity, and approval rules
• Track which engine version last operated each workspace
• Prevent production workspace mutation from development-state engines
• Run E2E evaluations before release or major workflow changes

The goal is not to let AI do everything autonomously. The goal is to give people and AI a reliable, auditable, and continuously improving way to work together.