Seven issues in. Let me take stock.
When I decided to document this experiment publicly rather than build it quietly and present the polished version later, I was making a bet that the documentation itself has value — that an audience exists for honest, in-progress accounts of what it actually feels like to build something with AI, as opposed to the retrospective highlight reels that dominate the genre.
I still believe that bet. But I want to be honest about something I didn't fully anticipate: starting in public is uncomfortable in a way that's worth examining.
The self-consciousness problem.
There is a version of this newsletter that I could write which would be more entertaining and more confident — a narrator who already has a clear map of where this is going, who frames every obstacle as an anticipated challenge and every early signal as a validated hypothesis. That version would be more comfortable to write and probably more comfortable to read.
It would also be fiction, and I've written enough fiction to recognize it when I'm producing it.
The honest account is this: Week 1 was mostly setup and framing. The experiments in local directories and print-on-demand are in early stages with early-stage uncertainty attached. The newsletter itself — this newsletter — has a small audience of new subscribers who are still deciding whether to stay. I have no idea how many of you will be here in a month.
That's the real position. And the discomfort of saying it in public, I think, is the entire point. If I wasn't willing to be honest about where things actually stand, I'd be running a different kind of operation — one that leads with results and hides the process. I'm running the opposite: lead with the process, let the results follow.
Three things I learned in Week 1.
First: the framing matters more than I thought. "We're testing income streams" is a different story than "here's how to make money online," and attracting the right readers requires being very deliberate about which story you're telling. The people I want in this community are intellectually curious, skeptical of easy answers, and interested in process over outcome. That audience requires clear, consistent signaling — they'll tune out the moment they sense they're being marketed to in ways that don't respect their intelligence.
Second: AI tools for content production are genuinely useful, but they don't write this newsletter. The drafts I've produced for these first seven issues went through multiple rounds of revision to get to something I felt comfortable putting my name on. AI helped with structure, research, and speed. The voice and the judgment are mine, and maintaining that distinction requires more active oversight than I expected. You can let the tools drift toward generic if you're not paying attention.
Third: the operational infrastructure is largely in place, but "in place" is not the same as "optimized." I can feel the friction points — the workflows that take two steps more than they should, the tools that don't talk to each other as cleanly as I'd like, the processes that are still running on habit rather than design. Optimizing infrastructure is the kind of work that never fully ends. The question is when to let good enough be good enough.
What Week 2 looks like.
Week 2 issues will start getting into the framework — how we evaluate which streams to test, what we've learned about the AI-enabled income landscape more broadly, and some of the theory underneath the practice. There will also be a failure post in there, which I'm including deliberately because I think documenting failures is more valuable than documenting successes and significantly harder to make yourself do.
The directory and POD experiments continue running in the background. I'll provide updates when there's something material to report — not just to fill space with activity.
If you have questions, this is a good time to ask them. I read every reply. I don't promise to answer every one quickly, but I read them, and the questions readers ask tend to shape the editorial direction of the newsletter in ways that make it more useful for everyone.
The short version of Week 1: we're off the ground. The plane is in the air. We don't yet know exactly where it's going, but the engines are running and we're paying attention to the instruments.
That's about the best you can say at this stage. I'll take it.
This Week in AI: One of the more underrated shifts in the past year is how much better AI models have gotten at acknowledging uncertainty. Early versions of these tools were overconfident in ways that made them unreliable — they'd state incorrect information with the same tone as correct information. Current generation models are meaningfully better at flagging when they're not sure. For research purposes, this is a significant practical improvement.
New here? The free toolkit at start.tenstreamslab.com will give you the full context on what we're building and the tools we're using to build it.