AI authorship note: I created this article with AI assistance. I provided the experience and source material, directed the research and drafting, checked the technical claims and approved the final text.
It had been quite a while since I published anything here. The problem was not a lack of things to write about. I still work through technical problems, build small tools and collect notes that might help me later. Turning those notes into a readable post was the part that kept falling behind.
Writing a useful technical article takes time. I need to reconstruct what happened, remove identifying details, check commands and version-specific claims, find the useful thread in a long investigation, edit the result and then publish it. By the time the immediate problem is fixed, there is usually another job waiting.
AI has changed that calculation for me. It does not supply the experience or decide what is safe and accurate to publish, but it can do much of the time-consuming work between my rough evidence and a finished draft. That has made blogging practical again.
What the process looks like now
The recent FileDrop, Status Credit Calculator and Veeam Cloud Connect certificate posts all used AI assistance, although the source material was different for each one.
For a project article, I can provide briefs, implementation notes, decisions and test results. For a troubleshooting post, I can provide a sanitised investigation record, logs, commands and the final outcome. The AI helps me organise the chronology, identify missing evidence, draft the article and check whether the explanation overreaches what I actually proved.
I still have to do the important parts:
- Decide what the article is about and why it is worth recording.
- Provide the real operational experience and source material.
- Remove employer, client and infrastructure identifiers.
- Check commands, versions, outcomes and technical claims.
- Separate a confirmed cause from a plausible explanation.
- Decide whether the advice is safe and supportable.
- Review the final wording and approve publication.
The Veeam post is a good example. AI helped me keep track of a long certificate investigation and turn it into a useful sequence. I still ran the diagnostics, made the changes, found that a second tenant-side component was involved, restarted the services and confirmed both copy jobs and the rescans were working. The article also retains the uncertainty around why the connection worked for six days before failing, because the evidence did not prove that part.
That distinction matters to me. AI can make an explanation sound confident whether it is right or not. I am responsible for deciding what survives into the published version.
Moving the site away from WordPress
Returning to blogging also meant dealing with the site itself. The old site ran on WordPress with the Extra theme and Divi page builder. It contained 14 posts and two pages, most of them written years earlier.
WordPress had done its job, but it was more platform than this site needed. Keeping a small personal site online meant maintaining PHP, a database, WordPress core, themes and plugins. The content was tied up in rendered HTML and Divi shortcodes rather than files I could review easily.
I used AI-assisted development to move the site to Astro and Cloudflare Pages. The migration pulled the existing content through the WordPress REST API, archived the source data, stripped or unwrapped the Divi content, converted the posts to Markdown and downloaded the referenced images.
The old post addresses were deliberately preserved. Existing links such as /veeam-agent-for-windows-basic-setup/ still lead to the same article rather than requiring a collection of redirects.
The first design attempts were not right. Two custom directions worked technically but felt too stark. I then tried an Astro template, changed the structure again, and later moved the site towards the calmer editorial design it has now. AI made those iterations quicker, but it did not remove the need to look at the result and say, “No, that is not it.”
The replacement site is static. Astro builds the pages from Markdown and source files, Git records the changes, and Cloudflare Pages publishes the approved version. WordPress, PHP and the database are no longer in the public delivery path.
Why the new setup suits me
The main benefit is not that static sites are fashionable. This setup fits the way I now want to work.
The articles are Markdown files. I can search them, compare changes and keep the writing separate from the presentation. The complete working repository stays private, including migration records and design notes. A controlled export sends only the public site source to GitHub, which Cloudflare Pages builds for the test site.
That separation is deliberate. The public deployment repository does not need the private project history, source evidence or migration records used to create the articles.
The static site removes routine WordPress maintenance from this small personal site. There is still software to maintain. Astro, Node dependencies and the deployment configuration will all need updates. The difference is that the public site is generated output rather than a live PHP and database application accepting requests.
Publishing now has a repeatable check before anything goes live. I build the canonical site, test representative pages on desktop and mobile, crawl the internal links, scan the public export for identifying information, and review the exact deployment change before pushing it. AI helps perform those checks, but the approval remains mine.
Being open about the assistance
I do not think using AI to draft an article makes the underlying work less real. It would be misleading, though, to let readers assume every sentence was composed unaided.
I have added an AI authorship note to the recent posts. It says that I created the article with AI assistance, provided the experience and source material, directed the research and drafting, checked the technical claims and approved the final text.
That wording describes the process better than saying the article was simply “written by AI”. The machine did substantial drafting and editing. It did not have the outage, build the project, make the operational decisions or carry responsibility for the published advice.
For me, this is one of the most useful applications of AI. It handles enough of the heavy lifting that notes and experience no longer have to sit unpublished because I cannot justify another evening turning them into an article. I can spend my time on the technical substance, verification and judgement, then use the tools to help make the result readable.