The eight families of AI workflows for technical writers

We have been researching the different AI documentation workflows technical writers use, as part of the development of our new Managing documentation projects with AI e-learning course. We posted our initial findings to a forum, and a few people responded with dazed eyes emoji. 😵‍💫

At first sight, the picture can look quite fragmented. Technical Authors are using a wide variety of workflows. But it is possible to identify some common themes.

In our research (we used ChatGPT to search the web and carry out an initial analysis), we found writers using AI to gather information: exploring codebases, creating release notes from API changes. Others were building scripts, testing documentation, or checking whether AI agents can actually use the docs.

They were using different tools and creating different outputs.

But having said that, we found that most of the workflows could be grouped into eight broad patterns, or, as ChatGPT calls them, families. This helped in understanding them from a perspective familiar to Technical Writers: there’s a beginning, middle, and an end. And within these processes, we could identify the roles assigned to people and to AI systems.

The eight families do not form a strict maturity model. There’s no perfect system. A highly automated workflow is not necessarily better than a simple one. For each organisation, the right approach will  depend on factors such as the task, the inputs, the risk, and how easy it is to verify the end result is accurate and trustworthy.

Alt text: Infographic titled “Eight Major AI Workflow Families for Technical Writers,” branded CherryLeaf. Eight vertical, color-coded columns show AI use cases progressing from lower autonomy on the left to higher autonomy on the right. Visible examples include: 1) AI as researcher or explainer—learning unfamiliar systems, interrogating code, summarizing APIs, identifying patterns, and preparing SME questions; 2) AI as outlining and planning partner—organizing notes and specifications, identifying themes and gaps, and drafting documentation plans; 3) AI-assisted authoring and editing—drafting, rewriting, applying style guides, transforming formats, and requiring human review and verification; and 8) Documentation for AI Readers—designing documentation so AI agents can find, understand, and use it, including improving information architecture, creating quick references, using llms.txt or product skills, and testing with real agents. A horizontal arrow along the bottom indicates increasing autonomy from “Lower autonomy (closer human involvement)” to “Higher autonomy (more automated, ongoing workflows).”

Here are four examples from the eight families:

AI as a researcher or explainer

In some workflows, AI is used primarily to help the Technical Writer understand something.

For example, it might help explain an unfamiliar codebase, identify patterns in source material, or suggest questions to ask a subject-matter expert.

A common pattern is:

  1. Source material
  2. AI explanation
  3. Writer checks understanding
  4. Better questions

This is close to what Fabrizio Ferri-Benedetti describes as the “watercooler” mode of AI-assisted technical writing. Here, AI is supporting the writer’s thinking rather than producing the final documentation.

AI as an outlining and planning partner

AI can also help turn raw information into structure.

Technical Authors/Writers use AI to analyse data to identify themes, possible gaps and potential documentation structures. That data could be meeting notes, specifications, tickets, and/or SME comments.

For example:

  1. Notes and specifications
  2. AI identifies themes
  3. Proposed outline
  4. Writer revises
  5. Documentation plan

The writer still makes the final decisions about audience, structure and scope.

AI-assisted authoring and editing

This is probably the most familiar type of workflow in use today. 

A Technical Writer might get an AI system to draft a first version, rewrite text, apply a style guide, transform content, or provide inline writing assistance.

A common pattern is:

  1. Trusted source material
  2. AI drafting or editing
  3. Human review
  4. Verification
  5. Publication

The quality of the result depends heavily on the quality of the source material and the verification process. The writer’s role as editor and verifier becomes increasingly important. 

Documentation for AI readers

Increasingly, AI systems are also becoming users of documentation.

For example, coding assistants and agents might need to retrieve technical documentation when they are writing code or troubleshooting problems.

Dachary Carey, Tom Johnson, Sarah Deaton, and others, have explored aspects of this. They are beginning to experiment with things such as agent-friendly information architecture, concise reference material, and systematic testing.

The other four families

The remaining families look at areas such as:

  • using code and diffs to drive documentation updates
  • using AI to build deterministic tooling and automation
  • using agents to maintain documentation
  • using AI and automated tests to verify documentation

Another viewpoint: autonomy

Alt text: Infographic titled “Low Autonomy to High Autonomy Automation.” A large arrow runs left to right, showing increasing AI autonomy. The subtitle explains that as autonomy increases, the human role shifts from direct writing to workflow design, governance, and review. Seven stages are shown in sequence: 1) Explain, 2) Suggest, 3) Draft, 4) Edit, 5) Execute, 6) Monitor, and 7) Repair. At the low-autonomy end, a callout says “Human closely involved in each step.” At the high-autonomy end, a callout says “Human sets rules, approves, and monitors.” The CherryLeaf logo appears at the bottom.

There is another useful way to look for common patterns: how much autonomy the AI has.

At one end, AI is closely supervised and acts mainly as a thinking or writing assistant. At the other, AI may be able to execute tasks, monitor documentation systems, propose repairs, or take action within defined rules.

As autonomy increases, the human role changes.

For some, it is moving from doing every step to designing the workflow, defining trusted sources, setting permissions, establishing review points, and monitoring results. In other words, the role moves from performing every task towards designing and governing the system that performs the tasks.

More information

The topic of workflows is one of the key themes in our Managing documentation projects with AI e-learning course.

The course explores all eight workflow families in more detail, along with practical examples, human review points, automation, risk and governance.

 

Images created using ChatGPT.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.