I built it twice. The fear changed shape the second time.

Fifth one, and the hardest to write.

Sonic Bloom is a website and fan-engagement platform for working musicians. One login instead of the five tools most of us cobble together, 0% commission on tips and sales instead of the cut Bandcamp takes. I built it because I run Free River Studios and a label, Tone Farmers, and I was tired of watching artists lose money to platforms that weren’t built for them. I’ve now built it twice.

The first time, in September 2025, I didn’t pace myself at all. For about a month I worked something close to full-time-and-a-half: mornings, evenings, and the hours before dawn, seven days a week. Somewhere around 90 hours a week, for weeks. In one sitting during that stretch I shipped fan signup, a tip jar, and a blog system together, over 6,000 lines in a single push. By mid-October I had a real multi-tenant platform, 230 passing tests, and a first real test with a real band. The Cullen Brothers, a friend’s group who agreed to be my guinea pigs, a QR code on their merch table pointing fans at their new page. I never went further than that. No public post, no wider ask. Just one friendly band, testing a favor.

Then I spent the next two and a half months rebuilding the same piece of it, twice, for two different reasons.

The first version handed artists a blank page and a shelf of components: build your own site, piece by piece. Testing it made the problem obvious fast: that’s a tool for someone who wants to build something. A working musician wants a finished website. I wrote it down at the time: artists don’t want a website builder, they want a website. Give them a house, then let them renovate. That wasn’t scope creep. It was a wrong first guess about what “self-serve” should mean, caught early enough to fix.

The second rebuild wasn’t a change of mind, it was a wall. Two duplicate services doing the same job, fighting each other. A model name that had quietly stopped existing. Menu items pointing at pages that 404’d. I’d fix one thing and something else would break, and at some point I just wrote, “why is it so hard to get anything done,” and meant it. That’s not a design problem. That’s a codebase that had accumulated too many ways of solving the same problem, and the only way through was to stop patching and rebuild the plumbing.

I quit on January 6th. There’s no dramatic entry for it. The record from that day is completely administrative: cancel Forge, destroy the droplets, downgrade Postmark, confirm GitHub has everything. I still don’t fully know why that was the day. My energy for this kind of thing comes and goes in waves, and I think I’d lost faith. Not in the code, but in whether the thing had real value or could compete out in the world. Two and a half months of “95% done, launching this week” that kept not being true didn’t help. Eventually I just stopped.

For six months the infrastructure sat dark, saving me about fifty bucks a month, while I built other things: other apps, other pieces of Warren Labs. And people kept asking me about it. That, as much as anything, is what got me to try again.

I came back to it in April, and the thing that had changed wasn’t my enthusiasm. It was the diagnosis. I said it to Claude plainly: I’d been building for months and hadn’t shipped to a single real user, and the website builder was the bottleneck every single time. Everything else already worked and people already liked it: the tips, the fan list, the show calendar, the link-in-bio page. The one thing eating every rebuild cycle was the piece nobody had asked for. So I killed it. Not patched, not simplified. Cut the component builder entirely, replaced it with fixed templates pulling from the same data, and put the ambitious page-editor version on the list of things to build later, once there was a business to build it for.

Everything downstream of that decision was different. One flat price, $20 a month, no tiers, down from a pricing maze that at one point had five different numbers on it. I validated the entire billing path in production, a real $20 charge clearing end to end, before I invited a single artist. I opened a separate business bank account before the first dollar came in, so revenue would never get tangled with anything else. And at some point I told Claude directly that I have a habit of stalling at 90% done, polishing things nobody’s asked for instead of shipping, and asked it to call that out when it saw it happening. It did, at least once. That’s not something I did the first time around.

Restart to a real, verified charge: about six and a half weeks. Not the two and a half months of almost-done that ate the first attempt.

Then it came time to actually tell people, and neither time did I really do it. The first attempt, I never posted anywhere. I just stopped, doubting the whole thing. The second time, the plan was personal invites first, a handful of artists I actually know, before any public post. I didn’t do that either. Every person I thought about reaching out to, I felt like I needed to have them over to the studio, walk them through it myself, watch them use it. And instead of doing that, even once, I skipped straight to a single Facebook post. No hands-on setup, no real artist walking around with a live site to point at. Which means the “real users” I could show off right now aren’t really real yet. Cullen Brothers is still sitting on their old profile from the first attempt, and I haven’t rebuilt it on the new version. That’s on me, and it’s probably the next thing to do.

Here’s the part worth sitting with. The first time, the fear was: does anyone want this, and can I even build it. I handled that fear by quietly stopping. The second time, with the building basically solved, a different fear took its place. Not “can I build this” but “can I ask someone I actually know to use this, and stand there while they do.” I handled that one by posting on Facebook instead, which is a real launch step but not the one I’d planned, and it let me avoid the part that scared me. A post doesn’t require anyone to say no to my face. I still don’t know if Sonic Bloom is something people want or something I’ve been circling because it’s more comfortable than finding out for sure. Most days I believe it’s real. Some days I don’t. That’s the current, unresolved state of it. Not a tidy lesson, just where I am.

This is also, plainly, why Sparkfile exists. Not as an abstract thesis about founder fallout, but as this, specifically: momentum that comes in waves when it’s just you, and a real risk of quietly stalling out on something worth building because the scary step is easier to skip than to name. I built a team of AI cofounders to teach other people how to not do what I just did twice. Turns out I needed to use it on myself.