Editorial queue and an agent behind a flag
A day of preparing several posts, adjusting publication dates, and making the public site agent more controlled with a feature flag.
The day was a mix of intense editorial production and product adjustment. On the content side, I moved a queue of posts about JavaScript for beginners, REST versus RPC, AI productivity, trust in code review, software development in 2026, agentic AI, and AI-assisted coding setup. Some texts became ready for publication. Others were kept as drafts or had publication dates adjusted.
The important part was keeping the queue organized. With several agents working in parallel, publishing is not only writing. It also means controlling published, date, image, translation, tone, and the moment each text should appear on the site.
Agent with an explicit boundary
On the product side of the site, I worked on the public agent. I centralized flags, put the feature behind a feature flag, added an environment example, and tightened fallback behavior. I also improved keyword matching so the agent avoids responses that are too generic.
That care matters because a public agent needs to be predictable. It cannot feel like a window into everything on the machine. It should answer inside the allowed corpus, fail honestly, and remain easy to turn off.
Content as agenda
The day's posts also worked as a study agenda. REST versus RPC brought up contracts, OpenAPI, stubs, and client ergonomics. The JavaScript post returned to fundamentals. The AI texts looked at productivity, anxiety, trust, review, and the changing software process.
There was also visual work on covers, especially for agentic themes. The image is not just decoration on this blog. It helps communicate the thesis before the first paragraph, but it still needs to stay light enough for Open Graph and social previews.
It was a day of treating publishing as a system: content, dates, flags, fallback, image, and judgment moving together.