The default solo-founder stack? I checked mine. I don’t have one.

A tweet made the rounds recently with a "default stack for AI-era solo founders": Claude for code, Supabase for the backend, Vercel for deploys, Stripe for payments, Resend for email, Clerk for auth, almost all of it on free tiers.

I sat down to write my own version and got stuck halfway. I run four products, and almost every one answers differently.

KeyEcho, the desktop app that gives your keyboard mechanical-switch sounds: one VPS behind Caddy, and the data is a single SQLite file — orders and license keys included. Backing it up means copying a file. There are no user accounts. You pay, you get a license key, and if you lose it, you recover it by email.

Reelloft, the local media library I'm finishing up, goes further: no server at all. The product is "no account, no cloud, no server." Payments run through a Stripe Payment Link, and activation doesn't need me online.

Broadside, the newsletter system I built — it's sending you this email. Its data sits in self-hosted Postgres, in a container I run myself.

difflore: deployed on Fly, Postgres on Neon, Redis on Upstash. It looks exactly like the stack in that tweet, managed all the way down.

Same person, same year: four products, four shapes of infrastructure, from no server at all to fully managed. Three are in production; Reelloft is nearing launch. The choice comes down to product shape: how much state there is, whether devices need to sync, and whether an hour of downtime hurts anyone. A single-machine desktop product gets a file for a database. Something that should scale to zero and never make me touch a machine hands its state to Neon. Anything in between, on a box I already keep running, gets a Postgres container.

Some pieces do carry across products. This week I put PostHog behind a same-domain reverse proxy to reduce data lost to common blockers, wrote it as a Caddy snippet, and made the next product's integration a one-line import. R2 handles product downloads through the same distribution pattern and tooling. What carries over is the pattern, not a platform.

And some services I just pay for: Stripe for payments, Resend for deliverability, Cloudflare for DDoS protection. If one of those providers goes down, no amount of AI-assisted log reading at 3 a.m. will bring it back.

Looking back, that tweet lists seven services but sells one promise: you only have to learn one stack. That promise used to be worth paying for. Every extra stack meant another set of docs to grind through, and another system only you could fix at 3 a.m.

AI compressed that cost. Evaluating options, wiring boilerplate, reading logs in the middle of the night — all of it got cheaper. With the "learn one stack" premium gone, one question decides what I keep: when this service breaks, can AI fix it before I finish a glass of water? If yes, I own it. If no, I pay.

For the first time, a one-person company doesn't need a default stack. Each product gets infrastructure shaped like the product.

How I pick the framework layer is in the last issue.

Reelloft is close. A product whose selling point is having no infrastructure — the next email will probably be its launch note.