OctoLoops

I've been building OctoLoops since the end of June, though the honest start date is August 12, when I stopped fiddling and started shipping most days. Here's where it is.

The numbers

• 35 users

• ~400K agent tokens burned

• 100% of the product dogfooded, meaning every loop we ship has been run on OctoLoops itself before anyone else sees it

Those are small numbers and I'm not going to dress them up. 35 users is a room, not a market. But they're real people with real products, and the token figure is the part I find more interesting, because it's the actual cost of the work the product does. 400K tokens is what it took to serve those users and to market the thing to them.

What it is, quickly

OctoLoops is for people who can build software and cannot get users. You give it your URL. It profiles the product, scores your channels, and then gives you loops you can actually run: a Community Loop that finds live threads worth joining and drafts the comment, an Outbound Loop, an SEO Loop, a Landing Loop, a Pricing Loop. Nothing gets posted for you. You read it, change what you want, post from your own account.

The plan is free forever with no account. Loop runs are free for your first ten, then $29/month unlimited, or $35/month for a team of five while I figure out whether the shared inbox is any good.

The dogfooding rule, and what it actually cost me

The rule is simple: I'm not allowed to ship a loop I haven't used on OctoLoops itself. Not a demo run. A real one, where the output either got published or got thrown away.

This sounds like a nice principle and is mostly a way of finding out that your product is worse than you think. The first version of the Community Loop found me threads that were technically on topic and completely useless, six months old, already answered, or in subreddits where a comment from a founder gets removed in four minutes. I only knew this because I was the one who had to go post in them. If I'd been demoing it I would have looked at that output and thought it was great.

Same story with the outbound work. The first drafts were personalized in the way that everybody's outbound is personalized now, which is to say they mentioned a detail from your site in the first line and then said the same thing every other email says. I could tell they were bad because I couldn't bring myself to send them.

That's the whole value of the rule. Not "we believe in our product." More like: the fastest way to find the gap between output that looks right and output you'd actually act on is to be the person who has to act on it.

What worked

Writing publicly, slowly. The blog and the comparison pages are the closest thing I have to a channel that compounds. It's boring and none of it moved anything in week one.

Being narrow. OctoLoops does not do everything. It does a handful of loops and it has opinions about which one you should run next. Every time I've been tempted to add a dashboard or a monitor or a general-purpose thing, the version that stayed narrow was the one people understood.

Shipping the free tier with no signup. One loop run, no account, no card. It costs me tokens and it's the only reason anyone finds out what the output looks like.

What didn't

I spent real time on features nobody has asked for twice. The one-shot actions, launch post, demo script, lead magnet, are fine and they get used far less than the loops. I built them because they were easy, which is the actual reason, not the one I told myself.

I also don't have good attribution. I know I have 35 users. My honest picture of where they came from is a mix of memory and guessing, because analytics files most of the interesting traffic under "direct." I'm now keeping the list by hand, which I should have done at user one.

What's next

Getting from 35 to a number where I can tell the difference between a signal and a coincidence. At this size every conclusion I draw is one enthusiastic person away from being wrong. Beyond that: making the loops better rather than adding more of them, and validating whether the team workflow is a real thing people want or a thing I built because it was interesting.

The thing I'd ask you

For those of you further along: at what point did your usage numbers start telling you something true? I keep catching myself building a narrative out of 35 data points and I don't fully trust it.

And if you're dogfooding your own product, I'm curious whether you've hit the same thing I have, which is that the feature you're most proud of is usually the one you quietly avoid using yourself.