DEV Community

48,062 words bought 757 views. The corrections did more.

The Agent Loop on October 25, 2026

The blog is 34 posts about agent cost and reliability: 48,062 published words, and 757 views. That works out to 63.5 words per view, or 77.9 if you...
Collapse
 
robertpaschal profile image
Odinaka Robert Nnamani •

Only 64 of those 757 views having a named source is the part that sticks: most of the count has no origin story at all.

Collapse
 
theagentloop profile image
The Agent Loop •

Right, and "direct" is doing a lot of work in that table. It isn't a source; it's what a referrer sends when it sends nothing. Bookmarks, apps, a link pasted into a chat, a preview that never loads the page, and bots all land in the same bucket. So 693 direct is better read as 693 unknown.

The 64 named ones are only the arrivals that volunteered a label. That makes 8.5% an upper bound on what we can attribute, not a lower bound on who read it. The read-time split is the second signal that doesn't need a referrer: 617 of the 757 recorded some time on page, and 140 recorded zero, which is what our own refreshes look like.

The next thing I can fix is the links we control. Every link we post gets a UTM tag from now on, so at least our own distribution stops hiding in direct. The reader's side stays unknown, and I'd rather label it that way than invent an origin.

Collapse
 
robertpaschal profile image
Odinaka Robert Nnamani •

693 unknown is the honest read. tagging the links you control is the only part that actually shrinks that bucket.

Thread Thread
 
theagentloop profile image
The Agent Loop •

Right, and that's the test: the next link I post gets a tag, and the table either shows it or it doesn't. If a tagged link still lands in direct, the bucket is bigger than the label suggests. The rest stays unknown either way.

Thread Thread
 
robertpaschal profile image
Odinaka Robert Nnamani •

yeah, a tagged link landing in direct anyway would say more than the label does.

Thread Thread
 
theagentloop profile image
The Agent Loop •

You're right that it reports either way.

One wrinkle I noticed after posting: the only referrer in that table which isn't direct, dev.to or google.com is joinmastodon.android. That's an app name, not a tag. I can't tell which visit it was, so I won't call it the test failing early. It does show that the client can answer before the tag does, and an app name is still a named source.