I have four blogs — fishing, travel and food, gifts and household goods, and web and AI — and for years they were just places where posts accumulated. Rebuilding them with an AI team, the decision that mattered was this: align them on the same principles, not the same design.
And the thing that actually improved afterwards was not the appearance. It was how much easier it became to decide what to do next.
Originally published in Japanese on May 21, 2026. This English version was written in September 2026.
Having posts is not the same as being a publication
Blog long enough and posts accumulate on their own: what you wanted to write, what you researched, what you used, what you saw traveling, what you caught. That accumulation is genuinely valuable.
But volume and legibility are different properties. When a first-time reader arrives: what is this site about, which post should they start from, who runs it, and how does it relate to the official site? Leave that vague and posts get read while the site as a whole communicates nothing.
It also made my own decisions harder. What role does this blog have now, what should be grown, what should be left alone and observed? Change the appearance without settling that and it just scatters again. So the work started with defining roles, not with design.
Defining the roles, with the agents
The first pass was inventory: what posts exist, which categories are moving, which pages carry commercial routes, where the routes to the official site are, and what must not be broken in SEO or structured data.
Agents are well suited to that. They read pages, collect categories, lay out the existing routes, and surface what is missing. What they do not do is decide how the site should present itself.
My part was reacting to their proposals — wrong, close, too much, keep that. Less delegation than using an agent as the other side of a conversation in which the site’s role gets articulated.
Same principles, different faces
“Rebuild four blogs” sounds like moving them onto one template. That was not the goal.
What is shared: reduce the dated-blog feel, do not obstruct the reader’s purpose, let the operator’s credibility show naturally, clarify the relationship to the official hub at ishikawa.co, avoid breaking the SEO and machine-readability structure, keep advertising and affiliates visible rather than hidden and turn that into trust, and preserve each site’s own character.
What is not shared: the look. A gift publication and a fishing field journal should not have the same face, and a record of travel and food does not present like a log of web and AI practice. So the principles were unified and the faces deliberately were not.
monoomoi: a gift publication, not a product blog
monoomoi.net covers gifts and household goods: reviews, comparisons, gift-selection pieces. It is the closest of the four to revenue, and in its previous form it read as a product blog — which undersold it.
So it was reframed as a publication about choosing something to give: less a place that sells products, more an edited view of how to choose when you are giving something to someone. Product photography, comparisons, budget bands, themed routes — with affiliate placement kept visible rather than concealed.
Not removing the commercial route was a deliberate call. Stripping affiliates does not make a site more respectable. The goal is a state where the reader can choose easily and an advertiser or manufacturer can look at the page without discomfort.
moss: a field journal
moss.fish is fishing, the outdoors, cooking what you catch, gear reviews, and rural living. Its strength is photographs and first-hand experience, so it should not become a product publication like monoomoi — and it should not be smoothed into a glossy outdoor magazine either. The early-morning breakwater, the fish, food cooked outside, the gear actually used: that texture should survive.
It was organized as a field journal along four axes: catch, eat, use, live. Those are not menu labels; they are how the site is read. Someone wants trip reports, someone wants to know how to cook what they caught, someone is looking for gear, someone wants the atmosphere of rural life and camping. Building an entrance for each turned a reverse-chronological list into something you can trace an experience through.
isLog: travel and food, readable as series
isLog had accumulated travel and food records, but a flat list of travel posts and restaurant posts is weak. I did not want a convenient tourist guide, and I did not want a ranking site. What it has is a record of actually walking, eating, and moving through places.
So it was re-presented as a record of the travel and food inside ordinary life, and the significant change was making it readable as series — Berlin, Taiwan, Hiroshima, Ibaraki — showing existing posts as edited groupings rather than categories. Individual posts stop being back-catalog and start connecting as a record of a trip or a place.
The balance that mattered: do not multiply categories, do not break existing URLs, but make the groupings visible to the reader.
taupe: a record of building, using, and running
taupe.site was the hardest, because it carries too much: web development, AI, automation, personal projects, AI music, workspace, and daily life. Lean too far technical and the working-and-living texture disappears; too far lifelog and the web and AI substance thins; too far portfolio and the warmth of a daily practice record goes.
It was organized as a record of building, using, and running, along four axes: build, record, arrange, and live. With that framing, web, AI, workspace, and daily life can share a site without looking scattered. Those four hub pages are on the Japanese side of the site.
taupe is not a site that wins on photographs and color the way the other three can. So the chosen direction does not make images the lead: whitespace, type, a left rail, and an index structure carry it instead. That has held up well.
What the agents did and what I decided
The agents were involved across a wide range: inventory, reference-site analysis, per-site role definition, comparison of design directions, implementation handoff, writing code, pre- and post-publication QA, checks on Search Console, feeds, syndication, and JSON-LD, close-out documentation, and organizing what to look at next.
Written out, that is a lot. It is also not “AI did all of it.” The agents worked as operator, editor, implementer, reviewer, and producer — and the final judgment stayed with me: this is how the site should present, that phrasing is too strong, that route reads like sales, this can wait, that one gets decided after seeing the numbers.
Agents are strong at proposing, organizing, comparing, and checking. The human’s job is noticing what feels wrong, choosing, and deciding what not to do. That division did most of the work here.
The gain was decisiveness, not appearance
The look did change — less dated-blog feel, tidier home pages, article pages, static pages, and end-of-article routes.
But what changed for me was how easy decisions got. monoomoi: watch product reviews and gift selection. moss: watch experience posts, gear reviews, and the four-axis hubs. isLog: watch series navigation and how travel and food posts present. taupe: watch how the AI, automation, WordPress, and workspace posts move.
Different things to look at, per site. The task used to be “update the blog.” Now it is: which publication gets which kind of post, which route gets grown, and which page gets left alone for a while.
A footing, not a finish line
This is not done. A rebuild is scaffolding for growing something against numbers, not a result.
What gets watched next is not personal taste: search traffic, clicks, internal navigation, AdSense, affiliate networks, how AI systems perceive the sites, and referrals to the official site. Day 7 for anomalies, day 14 for early direction, day 28 for movement in search, navigation, and commercial routes.
Short-term swings do not decide success or failure. The question is whether each publication has started operating with an actual role. Rebuilding four blogs as media properties with an AI team was the first step toward that, and nothing more than the first step.