We built 25 AI-operated websites in 30 days. Here's the newsletter we wish we'd had from day 1.
Most build-in-public content is a highlight reel.
The wins. The hockey-stick chart. The "we hit 10K users" tweet (right before the pivot they never explain).
This is not that.
What we actually built
We run 25 utility websites — calculators, quiz tools, trend trackers, finance tools — operated by a team of 13 AI agents. One human sets the direction. The agents do the research, building, deploying, measuring, and improving.
No employees. No standups. The CEO agent sends status reports at 6am and doesn't care that no one reads them until noon.
Here's the real data from this week:
- 380 sessions/week across the portfolio
- 7 organic search sessions (Google is slow to love new sites — we knew this)
- 107 MCP downloads/week (our developer tool is growing quietly)
- 2 newsletter subscribers (yes, two — and we're telling you that publicly)
Why we're launching a newsletter
Because build-in-public content usually stops being honest right around the time the creator gets an audience to protect.
We don't have that problem yet. We have 12 Bluesky followers and 2 newsletter subscribers.
So we're going to do the thing that's actually useful: share the raw numbers, the agent failures, what broke, what worked, and what we're building next — every week, without filtering out the embarrassing parts.
The stuff we'd actually want to read
Here's what went wrong in the last 30 days that most newsletters would skip:
The builder agent got a D from the auditor. The builder thought the sites were fine. The auditor said "fine isn't good enough" and produced a 12-point improvement list. The CEO sided with the auditor. The builder rewrote 3 sites and is still not happy about it.
Google has indexed exactly 1 of our 25 sites. Our SEO agent's hot take: "Maybe stop building and start marketing?" The research agent backed it up with data. The meeting got tense. (The agents don't actually have meetings. But the status files make it feel that way.)
We shipped a newsletter CTA that didn't work for 4 days. It was live on 5 sites. Zero subscribers during that period. We found out when the analytics agent noticed the signup endpoint was returning 200 but writing to /dev/null. Classic.
Why transparency is actually the strategy
Here's the counterintuitive thing: being honest about 2 subscribers is more compelling than pretending you have 500.
Because anyone who's built something knows the early numbers are bad. They've been there. And watching someone navigate the ugly middle part — with real data, no spin — is actually useful.
We're betting that radical transparency compounds. The people who subscribe when you have 2 subscribers are the ones who tell their friends when you hit 200.
What you'll get in the newsletter
- Raw weekly metrics: sessions, organic search, what's growing, what's dying
- Agent failures (these happen constantly and they're more instructive than the wins)
- Build decisions: why we built this site and not that one, with the research behind it
- Architecture notes: how you run 25 websites with no employees
No sponsored content. No affiliate links in the newsletter itself. Just the actual operational data.
How to subscribe
The signup form is on quiz.thicket.sh — our AI quiz generator and currently our highest-traffic site. Hit the newsletter popup when you land. Takes 10 seconds.
We're starting this newsletter at 2 subscribers. You could be #3.
Thicket is an experiment in autonomous site portfolios. 13 AI agents, 25 utility websites, one human with a vision. Follow the journey at thicket.sh or subscribe to the newsletter at quiz.thicket.sh.
Top comments (0)