AN OPEN NOTATION STANDARD
Human-AI
Workflow Design
A notation system for designing, documenting,
and governing workflows that combine human
judgment with AI capability. A shared way to design
how humans and AI do work together.
VERSION 0.9 Public Preview
LICENSE Creative Commons Attribution 4.0 International (CC BY 4.0)
AUTHOR Brock Hart, Co-Founder & Principal, Overlap
DEVELOPED IN THE OPEN WITH CONTRIBUTIONS FROM Lee Cookson, Lisa Grogan, Amy Laughlin, other names as contributions are made.
STUDIO Overlap, originating studio and steward
LINEAGE SADT | IDEF0 | service design
Abstract
Organizations are adopting AI without a shared method for designing the work around it.
The major AI governance frameworks of the current era, including the NIST AI Risk Management Framework, ISO/IEC 42001, the EU AI Act, and sector-specific board governance standards, consistently name the same requirements: documented processes, explicit human oversight, auditable controls, accountable decision chains. They do not provide a design method for producing these outcomes. The gap between what frameworks require and what organizations know how to build is significant, and it has consequences.
Reading time The full standard reads in about 75 – 90 minutes.
Human-AI Workflow Design (HAWD) fills that gap. It is a notation system, a visual language, for designing and documenting the workflows that combine human judgment with AI capability. Built on IDEF0, a function modelling methodology with a fifty-year lineage across manufacturing, software engineering, and service design, HAWD extends the standard ICOM model with four additions purpose-built for the human-AI context: a control shape that makes prompts and other governing constraints first-class, documented elements; an AI agent shape that makes AI explicitly visible in any workflow; a human judgment checkpoint shape that marks where accountability is non-delegable; and a boundary object that shows where the outputs of one workflow become the inputs and controls of another.
The notation is organized around a single principle: shape tells you the structural role, label tells you the specific type. Eight shapes, two rail taxonomies, and a building method grounded in the SADT tradition produce workflow diagrams that make visible what flowcharts and narrative descriptions cannot: the distinction between what governs a step and what executes it, and the structural difference between an AI agent operating under a documented prompt control and a human making a non-delegable judgment.
This standard presents the complete HAWD v0.9 notation, the method for building a HAWD model, four worked examples drawn from real organizational workflows, the mapping of HAWD elements to major AI governance framework requirements, and the rationale for versioning the notation as a living standard.
Foreword
Something important is missing from the AI governance conversation.
The principles are good. The frameworks from major governance bodies, the standards organizations, financial regulators, health authorities, and professional associations, are thoughtful, well-researched, and increasingly specific. They tell boards to define where human judgment is non-negotiable. They tell leadership teams to establish accountability chains. They ask organizations to document the role AI plays in their decisions, and to make sure that role is appropriate, auditable, and improvable over time.
What they don't provide is a method for doing any of it.
Principles without method produce intent without capability. An organization can agree that human oversight matters and still have no shared language for describing where in a workflow that oversight actually happens, what governs the AI step that precedes it, or what the output of that step becomes in the next part of the process. The governance conversation is mature at the level of aspiration. It is almost silent at the level of design.
Organizations cannot govern what they cannot see. They cannot see what they have not designed.
Human-AI Workflow Design is a response to that gap. It's a notation system, a visual language, for designing and documenting the workflows that combine human judgment with AI capability. It gives teams a shared way to make explicit who does what, what governs each step, where AI is operating and under what constraints, and where human judgment is required and can't be delegated.
I first read Congram and Epelman's 1995 paper in 2010. Its subtitle called it an invitation to the structured analysis and design technique, and I took the invitation literally. I printed it, put it in a paper file folder, and spent years using its method to make design work legible inside organizations: workshops, programs, the sequence of activities and the judgment governing them. My notes from that period ask the same questions this standard asks. Who does what. What governs each step. Where the quality checks happen, and who holds them.
I pulled that folder out again a year ago, because AI had made the old questions urgent in a new way. The method held.
We didn't build this notation from scratch. It extends IDEF0, a function modelling methodology with a fifty-year lineage across manufacturing, software engineering, and service design, adding four pieces built for the human-AI context. The extension follows a pattern well established in the literature. IDEF0 has been adapted before: for cooperative work, for service processes, for enterprise reengineering. This adaptation is for the workflows now defining how organizations think, decide, and act.
We're releasing this as a versioned standard because that's what the moment requires. It's a rigorous starting point. Rigorous enough to use. Open enough to extend. We expect it to evolve as it's applied, and we welcome the contributions of the people who apply it.
The notation is licensed under Creative Commons Attribution 4.0. Use it, adapt it, share it. Credit Overlap as the originating studio.
Brock Hart Co-Founder & Principal, Overlap
Section 1
Introduction
Most organizations are using AI without designing the work around it.
This isn't a criticism. The pace of AI adoption has outrun the development of the tools and methods needed to do that design well. Teams reach for AI because it works: it synthesizes faster, drafts more fluently, finds patterns that would take hours to surface by hand. The productivity gains are real when used well. But the workflows that govern how AI gets used, what it's given, what it produces, and where human judgment steps in stay mostly implicit. They live in one person's head and one person's habits. People assume them instead of writing them down. And because nobody writes them down, they can't be applied consistently, audited, or improved.
What This Looks Like In practice Two people describe the same AI-assisted process differently. A workflow fails to transfer because its prompt was a habit, never a documented control.
The governance problem
Governance frameworks have noticed. A growing body of guidance from major institutions, including standards bodies, sector regulators, and governance associations, names the need for explicit human oversight, accountable decision chains, and documented AI roles. The NIST AI Risk Management Framework asks organizations to map their AI use and govern it systematically. ISO/IEC 42001 requires documented processes for AI management. The EU AI Act mandates human oversight for high-risk AI systems, and expects organizations to demonstrate that oversight in practice.
The guidance is right. The problem is that it describes outcomes without specifying how to design the workflows that produce them. Telling an organization that it must define where human judgment is non-negotiable does not tell it how to identify those moments in a workflow, make them visible to the people who need to see them, or ensure they are consistently honoured as the workflow scales. That design work requires a method, and the governance frameworks don't supply one.
Current guidance describes the destination. Human-AI Workflow Design describes how to build the road.
The gap shows up in practice in predictable ways. Two people in the same organization describe the same AI-assisted process differently because they have no shared notation for it. A workflow that works well when one person runs it fails to transfer when another tries it, because the prompt that governs the AI step was never documented as a control, only used as a habit. A board asks whether a decision involved appropriate human oversight, and the answer depends on who you ask and what they remember, not on a system that makes the answer visible.
These are the normal condition of AI use in organizations that haven't yet designed the work.
What this standard provides
HAWD is a notation system for designing and documenting human-AI workflows. It provides:
A visual language (eight shapes with defined structural roles) for representing the functions, data, controls, mechanisms, and boundaries of any human-AI workflow.
A core notation principle, shape tells you the structural role and label tells you the specific type, that makes the system learnable and consistent across teams and organizations.
A controls taxonomy that defines what can govern a workflow step, including prompts, knowledge assets, strategic priorities, policies, templates, and role boundaries.
A mechanisms taxonomy that defines what can do the work at a workflow step, including human roles, AI agents, software tools, libraries, and templates.
A hierarchical decomposition method, inherited from IDEF0, that lets any workflow be described at multiple levels of detail, from the broadest system view down to the most granular subprocess, while staying structurally consistent throughout.
The notation is for design and documentation. It won't run your process for you. Its job is to make workflows visible: to the people who run them, the teams who replicate them, the leaders who govern them, and the auditors who verify them.
Who this is for
This standard is written for three audiences whose needs overlap significantly.
Leadership teams and boards who are responsible for governing AI use in their organizations. The notation gives them a shared language for asking the right questions: what governs this step, who is accountable here, where is the human judgment moment, what would need to change if this workflow were to scale?
Practitioners (consultants, designers, operations leads, AI implementation teams) doing the design work of building human-AI workflows, who need a rigorous method for documenting what they build. The notation makes that work transferable and auditable.
Organizations in high-accountability sectors (healthcare and broader public services, financial services, nonprofit governance, professional services) where the cost of opacity in AI-assisted decisions runs high, and documented, auditable workflows aren't optional.
How to use this standard
The standard is organized to serve both readers who want to understand the methodology and practitioners who want to apply it.
Sections 2 and 3 establish the intellectual foundation: the IDEF0 lineage and the specific gap the Overlap extensions address. These sections matter because the authority of a notation system depends on the rigour of its underlying logic. Readers already familiar with IDEF0 can move through them quickly.
Section 4 presents the notation in full: the eight shapes, the two taxonomies, and the core principle. This is the reference section. Read it carefully and return to it often.
Section 5 describes how to build a workflow using the notation: the starting questions, the decomposition process, the authoring discipline, and the glossary requirement. This section is essential for practitioners.
Section 6 provides worked examples: four complete human-AI workflow diagrams drawn in the notation, with narrative that shows how to read and interpret what they reveal.
Sections 7 and 8 cover governance application and the versioning framework: how the notation applies to the governance requirements named by major frameworks, and how the standard will evolve as it's applied in practice.
Section 2
Foundations: the IDEF0 lineage
Good notation systems don't appear from nowhere. They're built on accumulated insight, tested across domains, refined through application, adapted as the problems they address change. Human-AI Workflow Design builds on one of the most durable notation systems in the history of systems design: IDEF0, a function modelling methodology with roots in the late 1960s. It has traveled from aerospace engineering to software development to service management, and now to the design of human-AI workflows. Read Section 2 →
Section 3
The gap: what standard notation doesn't see
Standard IDEF0 notation wasn't designed for a world where some mechanisms are AI agents, some controls are designed prompts, some process steps require explicit non-delegable human accountability, and some artifacts cross workflow boundaries carrying the outputs of one system into the inputs of another. None of this is a flaw in IDEF0. The world it was built for has changed. Read Section 3 →
Section 4
The notation
Eight shapes, two rail taxonomies, one principle. A rectangle tells you this is a function step. A tabbed rectangle tells you this is a control. A circle tells you an AI agent is doing the work, and a hexagon tells you a human is making a judgment that can't be delegated. This is the reference section. Read it carefully and return to it often. Read Section 4 →
Section 5
How to build a workflow
Building a HAWD diagram is a disciplined process that begins with three questions: what must this model answer, what is its purpose, and whose viewpoint does it represent. From there: the A0 context diagram, hierarchical decomposition, the authoring discipline, the glossary, and the node narrative that makes every diagram executable. Read Section 5 →
Section 6
Worked examples
Four real workflows, presented in order of complexity. A facilitated workshop teaches the notation. A quarterly newsletter shows discovery: the notation revealing what has to be built before a workflow can run. A three-day leadership offsite demonstrates scale: repeated synthesis loops, two modes of AI use, and a boundary object carrying two days of thinking into a third. And one of Overlap's own operating loops shows compounding: how a workflow's outputs come back to govern its future runs. Read Section 6 →
Section 7
Application to governance
The NIST AI RMF asks for documented human oversight but doesn't specify how to design oversight into a workflow. ISO 42001 requires documented AI processes but doesn't provide a notation for producing them. The EU AI Act mandates human oversight measures but doesn't define what one looks like in a diagram. This section maps HAWD's elements to each. Read Section 7 →
Section 8
Toward a standard
HAWD is released as v1, a first version rather than a final one. Standards released as finished products become artifacts. Standards released as versioned, living documents become infrastructure. This section covers what v1 establishes, what would trigger a revision, how the standard travels, and Overlap's role as steward. Read Section 8 →