What we removed
The site you are reading replaced one that had been up for about a year. Rather than write about what the new one does, it is more useful to write down what was taken out, because the removals are the argument.
The headline was made of adjectives
The old homepage said Transform request to action, in a pink-to-orange gradient with a glow behind it, above a rotating list of verbs. Further down it asked Ready to Transform Your Business? and offered 24/7 AI support · 14-day free trial · No credit card required.
Not one of those claims is checkable. That is the problem with them, and it is not a problem of taste. Any language model will write that paragraph for free, in any voice you ask for, about any company. A sentence that a machine can generate on demand carries no information about whether the thing behind it is real.
So the rule for the new site is that the primary content has to be facts that would be false if we stopped doing the work. The register on the homepage is the whole idea: what exists, when it started, when it last moved. Those columns cannot be written by a copywriter, and they go stale in public if the work stops.
There is no uptime column, because nothing here is monitored yet. That felt worth leaving out rather than filling in.
The site refused to load on a phone
There was a component called ScreenSizeGuard. Below 768 pixels it covered the page
with a modal that said Better Experience Awaits and asked you to come back on a
desktop.
It had a Continue Anyway button. The cookie banner rendered on top of it and covered that button. So on a phone the site was not merely discouraging, it was unusable, and it had been that way for every mobile visitor for a year.
That is deleted. Nothing replaced it.
The one genuinely useful page was invisible
The old site had an agent directory — a page per integration, describing what an assistant could do with Jira or Confluence or GitHub. The Jira page carried four hundred words of specific, accurate detail. It was the best content on the site.
It was fetched from a CMS in the browser. Which means this was what a crawler received:
$ curl -s https://devclusterai.com/agent-dictionary/ | grep -io "jira\|loading"
Loading
No major AI crawler runs JavaScript. GPTBot, ClaudeBot, PerplexityBot and CCBot each
make one request, read the HTML that comes back, and leave. So the index was blank to
all of them — and because the index was the only thing linking to the detail pages,
those pages had no inbound links at all. They were orphans. Every page also shipped the
same <title>DevCluster AI</title>, so even where they were found, there was nothing to
tell them apart.
The product those pages documented is no longer running, so they are gone rather than fixed. But the lesson survives the product: content that only exists after JavaScript runs does not exist.
Three fonts, none of them from a CDN
The old stylesheet opened with three @import calls to Google Fonts. That is a
render-blocking dependency on somebody else's uptime, and it hands every visitor's IP
address to a third party before a single word appears.
The new site uses Nimbus Sans Narrow, Century Schoolbook and Hack — subset to the characters actually used, self-hosted, 115 KB for six faces. They were chosen because they belong to the subject: Nimbus Sans Narrow is Helvetica Narrow, the lettering of technical drawings, and Century Schoolbook is the face of legal briefs. A register should look like a register.
What is left
Four projects, one of them public, a small team, and a changelog. That is a smaller claim than the old site made. It has the advantage of being true, and of getting less true if we stop.