Cloning takes a weekend, finding the 38% took months — Chovy's Blog

Cloning takes a weekend, finding the 38% took months

2026-08-19, by Anthony “chovy” Ettinger.

How this was written: drafted with an AI assistant from my own notes,<br>then edited by me.

I posted this on<br>LinkedIn<br>yesterday:

Hire us or we'll clone your ass in a weekend. $150/hour on a c2c.

It reads like bravado. It is closer to a complaint, and the complaint is about where the value<br>went.

The cloning part is true and boring

Pick a SaaS product. Auth, billing, a dashboard, a table with filters, a settings page, an API<br>with keys. A competent person with an agent can have that standing up over a weekend. I am not<br>bragging about a skill here. Everybody who does this work full time knows it, and the ones selling<br>six month rebuild engagements know it best of all.

So if the software is the moat, there is no moat. That part of the pitch is a threat you can<br>verify yourself in two days.

What a weekend does not buy

Here is what I actually did yesterday, on my own product, and it is the argument.

RSS Amplifier is a feed directory. It carries sponsored items in the feeds it publishes, one per<br>ten entries, three at most, never last. That rule shipped weeks ago and looked fine. Yesterday I<br>went looking at why one feed had no ads in it, and found that the rule places an ad after every<br>tenth post and refuses to let one sit at the end, so a feed with exactly ten posts has its only slot<br>at the very end and drops it.

Ten is the number an RSS document conventionally carries. In that directory, 32,114 of the 83,940<br>feeds with items carry exactly ten. So 38% of the inventory had been unsellable since the day the<br>feature shipped, by arithmetic, with no error anywhere and every test passing.

Later the same day, the author enrichment. It reads each blog to work out who writes it. One<br>blog on GitHub Pages published no rel=me, no h-card, and one outbound link, to the theme its author<br>used. The pass fetched the page, correctly found nobody, and moved on. The author's GitHub<br>account was sitting in the hostname the feed is served from. Ten out of ten of those feeds went from<br>naming nobody to naming a real person with two to seven ways to contact them, off a rule that costs<br>no network request at all because it is arithmetic on a URL.

You cannot clone either of those. They are not features. They are things that only exist after<br>somebody runs the thing against the real web for months and then goes and looks at what it actually<br>did, rather than at what it was supposed to do.

The pitch, said plainly

The weekend clone gets you the features. It does not get you the person who notices that a<br>default of ten is eating 38% of your revenue, or that the identity you need is in the hostname you<br>already have.

That gap is the whole business. It is why the rate is what it is, and it is why I keep writing<br>down the mistakes in public rather than the finished screenshots. The finished screenshot is the<br>part anybody can copy by Sunday.

Find me: Mastodon &middot;<br>GitHub &middot;<br>email