I use AI as part of the editorial process for some articles on this blog.

That does not mean handing over a topic and publishing whatever comes back. The process starts from my own experience, technical work, opinions, or questions. AI can then help at different stages, depending on the article.

I use it to make the final article clearer, better structured, and easier to read, not to avoid the work behind it. In practice, it often shifts effort away from turning rough thoughts into polished prose and toward deciding what the article should say, challenging the argument, and reviewing the result.

I remain responsible for the ideas, technical accuracy, examples, opinions, and final text that I publish.

How I use AI

The exact workflow varies, but it can include:

  • Sparring — challenging an idea, looking for weak assumptions, and finding a sharper angle before writing.
  • Interviewing — extracting the firsthand material behind the idea: what I actually did, why I made certain choices, what alternatives I rejected, and which trade-offs or constraints matter.
  • Outlining — organizing that material into a structure that is easier to write from without inventing missing reasoning.
  • Drafting and editing — helping turn rough notes or a conversation into prose, or improving wording and structure in an existing draft.
  • Review — reviewing a draft for clarity, structure, technical accuracy, tone, and publishability.
  • Media — creating or assisting with supporting material such as images or diagrams.

I deliberately keep these responsibilities separate. Rather than using one generic prompt for the whole process, I maintain a set of specialized blog-writing skills for sparring, interviewing, outlining, drafting conventions, review, and other parts of the workflow. A workflow skill coordinates them, but each stage has a different job and different constraints.

The separation matters because it is meant to prevent the model from filling gaps with plausible generic material. The interview stage, for example, exists specifically to extract my own examples, reasoning, terminology, and trade-offs before an outline is created.

Once a draft reaches a pull request, I can also ask a dedicated GitHub Copilot blog-reviewer agent to review it independently and leave feedback directly on the pull request. That review is another input, not an authority: I still decide which comments are useful, which changes to make, and which suggestions to reject.

I do not treat AI output as authoritative. Suggestions can be wrong, generic, or simply not sound like me. I reject or rewrite them when they do not fit the article.

AI assistance labels

When AI materially contributes to an article, I may disclose that using one or more of these labels:

  • Editorial — AI participated broadly in shaping the article, for example through sparring, interviewing, outlining, drafting, or iterative refinement.
  • Editing — AI materially helped rewrite, polish, or restructure prose that was already drafted.
  • Review — AI was used to review the article after or during writing.
  • Media — AI was used to create or assist with supporting material such as images or diagrams.

These labels are not meant to measure how much of an article was “written by AI”. They describe where AI participated in the process.

Responsibility stays with me

AI can make parts of writing faster, especially turning rough thoughts into clear prose and providing another pass over a draft. It does not remove the need to decide what the article should say, verify technical claims, reject weak suggestions, or read the final result carefully.

The publish button is still mine.