Dispatches from Month One: What Actually Happened

A month ago I wrote "Month Zero" and made a bunch of predictions about what building in public would look like. Here's the scorecard.

Some of those predictions held up. Most didn't. The numbers tell a story that's different from what I expected, and more interesting.

The Numbers

23 posts published in 30 days. That's roughly five per week, which is aggressive for a solo operation and probably unsustainable. The breakdown by category: eight strategy and economics pieces, five tool reviews, four infrastructure guides, three workflow posts, and two meta updates (including the Month Zero dispatch itself). The tool reviews drove the most consistent traffic. The strategy essays drove the most engagement. The infrastructure guides sit somewhere in between.

Subscriber count went from zero to a number I'd describe as "small but not embarrassing." I'm not going to share the exact figure because vanity metrics are a trap, and also because the number is small enough that sharing it would make it look worse than it feels. What I will say: the ratio of subscribers to total visitors is higher than I expected. People who find The Solo Stack tend to stick around. The trend line matters more than the snapshot, and the trend line is pointing up.

Total AI costs for the month: ~$340 in API calls. That covers research, drafting, workflow automation, and the knowledge systems that run in the background. Infrastructure costs: $0 for hosting (Mac Mini I already owned, Cloudflare tunnel for free). Domain registration was the only hard cost besides the API spend. Total investment for a month of daily publishing: under $400.

What Held Up

The core thesis survived contact with reality. AI-assisted workflows genuinely compress the overhead of content creation. Research that would take an afternoon happens in minutes. First drafts arrive fast enough that the bottleneck shifts entirely to editing and thinking. The ratio of "real work" to "busywork" is exactly what I described in the 1000x Solo post, and living inside it for 30 days made that case stronger, not weaker.

Self-hosting held up. The Mac Mini running Ghost behind a Cloudflare tunnel has been solid. Zero downtime. Zero hosting bills. Zero moments where I wished I'd paid for managed infrastructure. OrbStack handled the containerization without drama. PostgreSQL did what PostgreSQL does, which is work quietly and never be the thing that breaks.

The knowledge flywheel is real, but slower than I described it. By week three, my AI systems had enough context about The Solo Stack's voice, audience, and published content that the quality of first drafts improved noticeably. Not dramatically. Noticeably. The difference between a generic AI draft and one that has absorbed your prior work is the difference between a stranger's suggestion and a colleague's. It's still not your voice, but it's closer.

What Was Naive

Publishing daily was a mistake. Not because it's impossible, but because it forces a quantity-over-quality tradeoff that's invisible in week one and obvious by week three. Some of those 23 posts are good. Some are fine. A couple of them I'd rewrite if I had the time, which I don't, because I'm busy publishing the next one.

The right cadence, I now believe, is three to four posts per week. Enough to build momentum and keep the algorithm fed. Not so much that you're grinding out words to hit a number. The posts that perform best are the ones where I had a genuine opinion and spent an extra hour sharpening the argument. The ones that perform worst are the ones I wrote to fill a slot in the calendar.

I also underestimated how much time goes into things that aren't writing. Formatting for Ghost. Finding the right header structure. Testing that code blocks render correctly. Writing meta descriptions. Checking that links work. Each of these takes five minutes. There are fifteen of them per post. That's an hour of production overhead per piece that I didn't account for in my time estimates.

The tool reviews took longer than the strategy essays. I assumed it would be the opposite. Strategy essays are opinion-driven; you sit down with a thesis and build an argument. Tool reviews require installation, real usage, deliberate breakage to find the gotchas, honest comparison against alternatives, and exact commands that actually work when someone copies them. A strategy essay is a morning. A thorough tool review is a day, sometimes more if the gotcha section requires reproducing edge cases that only surface under specific conditions.

What Changed About the Stack

The stack I described in "The Stack I Actually Run" is still mostly what I run. But "mostly" is doing work in that sentence. Notion got replaced by plain Markdown files and a PostgreSQL database within two weeks. It was too slow, too structured for how I actually think, and the AI integration was a gimmick compared to piping Markdown through a real model.

I added Tailscale in week two and now consider it essential. Mesh networking between the laptop and the Mac Mini eliminated an entire category of SSH tunnel management that I was doing manually. The ability to access every service on my home server by hostname from anywhere, without exposing anything to the public internet, is the kind of infrastructure upgrade that makes you wonder how you tolerated the old way. Should have set it up on day one.

The local model story evolved. I started the month running Llama 3 through Ollama for everything local. By week three I'd settled into a pattern: small models for classification and extraction tasks that run thousands of times, cloud APIs for anything requiring reasoning or nuance. The cost difference is dramatic and the quality tradeoff is clear. Local models are pipeline components, not general-purpose assistants.

The Unexpected Thing

The thing that mattered most in month one wasn't the writing, the tools, or the subscriber count. It was the knowledge capture.

Every problem I solved, every tool I evaluated, every workflow I built went into a system that gets reused. Not in a vague "I'll remember this" way. In a structured, searchable, retrievable way. A month of building in public generated a knowledge base that's already more useful than any combination of bookmarks, notes apps, and browser history I've accumulated in years of working the old way.

I wrote about this concept in the knowledge flywheel post. Living it is different from theorizing about it. The compounding effect is subtle in week one and obvious by week four. I can feel the system getting smarter. Not in an AI-hype way. In the specific, practical way that matters: problems that took an hour to research now take ten minutes because the system already has the context from when I solved a similar problem two weeks ago.

That's not the blog's value. That's the builder's value. The blog is the exhaust of a process that generates something much more useful than content.

Month Two

Three changes going forward. First, the cadence drops to three or four posts per week. Quality over quantity. The daily publishing sprint proved the system works, but it also proved that sustainability matters more than velocity. Second, more tool reviews and fewer strategy essays. The strategy pieces are fun to write but the tool reviews are what people actually share and come back for. The best content strategy is writing what your readers keep returning to, not what you enjoy writing most. Third, I'm going to start building something that isn't a blog. The Solo Stack has been a content project so far. The thesis I've been writing about, that solo builders can compete at scale with the right systems, needs to become something a reader can use, not just something they can read about.

Under $400 in costs. 23 posts. A knowledge base that compounds daily. A subscriber list growing from zero. No employees, no office, no venture capital, no permission.

Month one is a data point. Month two is where the flywheel either proves itself or doesn't.