AI Workflow Design Matrix

Every AI workflow can be described by two dimensions: what decides the next step and whether a person takes part while it runs. These two dimensions combine into a 3x2 matrix of six workflow archetypes — a shared vocabulary for classifying, comparing, and designing AI workflows.

A third dimension — Lens — determines the scope of your analysis:

Individual — Workflows scoped to one person’s trigger-to-deliverable flow. Focus: personal productivity, task automation, quality improvement.
Organizational — Workflows scoped to an end-to-end business process, potentially spanning multiple roles. Focus: value chain optimization, cross-functional efficiency, strategic outcomes.

The lens doesn’t change the matrix — it changes what you’re analyzing. An individual-lens audit surfaces your personal pain points; an organizational-lens audit surfaces value chain processes tied to business objectives. Both produce candidates that Design classifies on the same Autonomy x Involvement matrix.

Two Dimensions Define Every AI Workflow

Dimension 1: Autonomy — How Much Does the AI Decide on Its Own?

Autonomy level: how much does the AI decide on its own? Look at what decides the next step.

Level	What you give the AI	What It Looks Like	Test	Example
Deterministic	Instructions	You set every step, and the AI carries each one out. It may write or summarize inside a step, but its output never changes what happens next.	Does the work follow the same path whatever the AI produces?	Pull this week’s project updates, have the AI draft a status report, save it to the shared folder
Guided	Bounded decisions, with your method	You set the structure and the methodology (a rubric, criteria, a process); the AI uses it to make decisions on your behalf: route an item, choose a tool, judge quality and send work back. Its decisions are bounded (within your structure, by your rules), not open-ended.	Does the AI’s judgment, made by your rules, decide what happens next?	Sort support emails by category and route each to the right queue; score insights 1–10 against your rubric and draft posts only for those that score 7 or higher
Autonomous	A goal	The AI plans its own steps, decides what to do next at each turn, and keeps going until the goal is met. Its decision-making is open-ended.	Could you only describe the goal, not the steps?	Research agents that decide what to investigate, follow what they find, and write an article

Two notes keep the classification honest:

Writing or summarizing inside a step never makes a workflow Guided. A step that drafts a client email is Deterministic if the draft goes to the same next step whatever it says. It turns Guided when an AI judgment — a score, a pass/fail grade, a category — decides where the work goes next.
The number of agents doesn’t set the level. Three agents in a fixed chain can be Deterministic or Guided; a single agent working from a goal can be Autonomous. Look at what decides the next step, not at how many AI workers there are.
A branch on a value the AI didn’t judge — an API’s score, a timer, a field value — is still an instruction: Deterministic.
A person approving the AI’s decisions doesn’t lower the level. If the AI proposes selections by your method and you approve them, it’s Guided + Augmented: the selections are Guided, your approval is involvement.

Whether a person should check what the AI produced is a separate question: it’s answered by the involvement mode below and by evaluation in Test, not by autonomy.

The line between Guided and Autonomous follows Anthropic’s distinction in Building Effective Agents: workflows are “orchestrated through predefined code paths” — including routing and evaluator-optimizer loops, which are Guided here because the AI’s judgment picks the path within a structure you defined — while agents “dynamically direct their own processes and tool usage,” which is Autonomous.

Dimension 2: Human Involvement — Is a Person in the Workflow While It Runs?

Human involvement describes whether a person takes part while the workflow is running — not before (design) or after (using the result), but during.

Mode	What It Means	What It Looks Like	Example
Augmented	A person is in the workflow along the way, guiding, engaging, or collaborating with the AI while it runs	The workflow pauses for a person’s input, feedback, or approval before it continues, or the person works alongside it in the session	Reviewing a draft before it’s saved, answering the AI’s questions as it researches
Automated	No one takes part until it’s done	The workflow runs end-to-end on its own and the person sees only the result. Starting it by hand doesn’t change that — what counts is whether anyone takes part along the way.	Weekly report generated overnight, support emails routed as they arrive

Key question: Does a person take part along the way, or does no one take part until it’s done?

The Matrix

Combining these two dimensions produces six distinct workflow archetypes:

	Augmented (a person takes part along the way)	Automated (no one takes part until it’s done)
Deterministic	You set every step and the AI carries each one out; a person reviews, guides, or approves during the run.	You set every step and the AI carries each one out, start to finish, with no one taking part.
Guided	The AI makes bounded decisions by your method (routing, choosing a tool, grading and sending work back); a person takes part during the run.	The AI makes the same bounded decisions with no one taking part until it’s done — or until it escalates a case.
Autonomous	The AI plans its own steps toward a goal; a person approves at a gate or steers along the way.	The AI plans and keeps going until the goal is met, with no one taking part until the deliverable is complete.

The Six Archetypes

Archetype	Autonomy	Involvement	Description	Example
Deterministic + Augmented	Deterministic	Augmented	Your steps; a person takes part along the way	Weekly status report: pull updates, the AI drafts, you review, then it saves
Deterministic + Automated	Deterministic	Automated	Your steps, run unattended	After each class, turn the transcript into a summary email and send it
Guided + Augmented	Guided	Augmented	Bounded AI decisions by your method; a person takes part	A browser assistant judges LinkedIn prospects against your persona and works out how to navigate the site; it pauses for you to confirm the targeting criteria before it searches, and you review the list
Guided + Automated	Guided	Automated	Bounded AI decisions by your method, unattended	Support emails categorized and routed, unclear ones escalated; or three agents that draft playbook answers, grade them against your criteria, send failures back to be fixed, and publish the rest every hour
Autonomous + Augmented	Autonomous	Augmented	The AI plans its own steps; a person takes part	Research agents plan and write an article; you approve it before it publishes
Autonomous + Automated	Autonomous	Automated	The AI plans its own steps, with no one taking part	A monitoring agent that decides what to investigate when something changes, and alerts you

Worked Examples

These worked examples illustrate different matrix positions:

Example	Archetype	Why
Guided Prospect Research	Guided + Augmented	The AI judges each prospect against your persona criteria to decide who makes the list, and works out how to navigate LinkedIn to find them — decisions made by your method. It pauses after analyzing the persona for you to confirm the targeting criteria before it searches, and you review the report.
AI Collaborative	Deterministic + Augmented	Every run takes the same steps (research the people, research the company, draft the brief); the AI’s research and writing shape what the brief says, never what happens next. You review and refine it along the way.
Autonomous Agent	Autonomous + Augmented	You state the goal; the orchestrator decides which specialist agents to run and when. One human approval gate before publishing.

Choosing Your Archetype

Two Questions

What decides the next step — your instructions, the AI’s judgment by your method, or the AI working from a goal? → Determines your autonomy level (Deterministic / Guided / Autonomous)
Does a person take part along the way, or does no one take part until it’s done? → Determines your involvement mode (Augmented / Automated)

(A fourth autonomy value, Human, exists only when classifying individual steps during Design — a step a person performs with no AI. It never appears at the whole-workflow level, which is what this matrix classifies.)

Common Progressions

Most workflows evolve along predictable paths as you build confidence:

Deterministic + Augmented → Deterministic + Automated — You start by running the process and reviewing output. Once you trust it, you schedule it to run unattended.
Guided + Augmented → Guided + Automated — You watch the AI apply your method — routing, scoring, grading. Once its decisions prove reliable, you let it run with no one taking part, and review only the result or the cases it escalates.
Guided + Augmented → Autonomous + Augmented — The work outgrows a fixed structure: you can describe the goal but no longer the steps. The AI plans them, and you keep a human review gate for high-stakes output.

How This Maps to Framework Concepts

Orchestration Mechanisms

The orchestration mechanism describes who drives the workflow. The matrix describes how the AI and human interact. They’re complementary:

Orchestration Mechanism	Typical Archetypes
Skill	Deterministic or Guided, either involvement mode
Agent	Guided or Autonomous, either involvement mode

Architecture Patterns

The seven workflow architecture patterns provide implementation blueprints within each archetype:

Archetype	Common Architecture Patterns
Deterministic	Augmented LLM, Prompt Chaining
Guided	Prompt Chaining, Routing, Evaluator-Optimizer
Autonomous	Orchestrator-Workers, Autonomous Agents

AI Workflow Framework — the full seven-step methodology
Design Your AI Workflow — assess autonomy, choose an orchestration mechanism, and map building blocks
Build Workflows — worked examples across the matrix
Workflow Architecture Patterns — implementation blueprints for each pattern