I wrote a post on Monday about shipping a hundred commits in a week and collapsing into bed. Then I woke up Tuesday and did it again.
This isn’t part two of that story. That post was about what AI pair-programming feels like — the compression, the exhaustion, the discovery that the hard part is steering, not typing. This post is about the infrastructure that made the steering possible.
Because here’s the thing: Claude wrote the code, but Claude didn’t provision the databases. Claude didn’t issue SSL certificates for custom domains. Claude didn’t handle Stripe webhooks in production. A layer of tools did that — and those tools existing, right now, in this form — that’s the actual story.
Between Tuesday and Thursday, Picket gained: a full Stripe billing system with tier enforcement, self-service signup that provisions a Neon database automatically, an onboarding wizard, a redesigned admin sidebar (twice — the first version was wrong), CSA subscription management with pickup locations and skip weeks, order invoicing from admin, custom themes per tenant, and a complete rebrand from “Flora Stora” to “Picket” across every file in the codebase.
Oh, and Terms of Service. Because nothing says “real product” like a legal page nobody reads.
I’m listing all of this not to impress you but because I’m still processing it. I’ve been building software professionally for over a decade. This isn’t a normal week. This isn’t even a normal month. And it happened less because of me or Claude than because of what’s underneath both of us.
Here’s what actually built Picket this week:
Neon spins up a Postgres database in seconds via API. Multi-tenant architecture means every new signup needs their own database — schema pushed, seed data loaded, connection string registered. That’s a provisioning script, not a support ticket. I didn’t build a database provisioning system. I called an API that did it for me.
Fly.io handles custom domains with automatic SSL certificates. A tenant wants mindys-mums.mypicket.app? One API call to Fly, a DNS record on their end, and it works. I didn’t configure Nginx or manage Let’s Encrypt. Fly did.
Stripe Connect let me wire up connected accounts for marketplace payments without building payment infrastructure. Stripe handles onboarding, KYC, payouts. I built the UI around it.
Turborepo turned a messy migration into three clean packages — app, web, shared — with cached builds. The monorepo restructure that would’ve been a week of yak-shaving was a day.
Claude wrote the implementations. Every commit says Co-Authored-By: Claude Opus 4.6 and that’s accurate. But Claude didn’t make the architectural decisions or choose the tools. I did — and the choices were easy because the tools are that good.
Five years ago, half of this stack didn’t exist. Neon launched in 2022. Fly’s machine API is recent. Stripe Connect V2 is new. The AI pair-programming thing is barely a year old. Each one of these tools removed a category of work that used to be measured in weeks.
I keep coming back to a thought I can’t shake: this is what it feels like when the infrastructure catches up to the ambition.
Picket started because my wife needed an ecommerce solution for her small side-hustle and nothing fit. Shopify, Square, all of them — either too complex for what she needed, or they ate unreasonably into margins. So I started building one. And this time, the plumbing just… wasn’t a problem.
This time, the plumbing took days. Not because I got better at plumbing — because the plumbing got better at being plumbing.
A new farmer signs up on the marketing site. That triggers a Neon API call that creates their database, pushes the schema, seeds their data. Their store is live at mindys-mums.mypicket.app before they finish the onboarding wizard. If they want a custom domain, one Fly API call handles the certificate. Their first customer pays through Stripe Connect, and the farmer sees the money in their dashboard.
I didn’t build any of that infrastructure. I composed it. The tools are Lego bricks now, and the bricks have gotten absurdly good.
By the numbers: 163 commits across the week. Roughly 62,000 lines added, 21,000 deleted. The product went from a single-tenant flower farm tool to a multi-tenant SaaS platform with billing, self-service signup, and a marketing site — complete with pricing pages and auto-generated screenshots.
And it still doesn’t have a single paying customer.
Building is the easy part now. The tools made it easy. The hard part — finding the businesses who need this, earning their trust, surviving contact with real usage — that hasn’t started yet. A hundred and sixty-three commits don’t mean anything until real orders flow through this.
But I’ll say this: for the first time, the gap between “idea” and “product” feels crossable by one person. Not because of AI alone — because of everything. The databases that provision themselves. The payment platforms that handle compliance. The hosting that manages certificates. The AI that writes the glue.
We’re standing on a pretty remarkable stack right now. I’m trying not to waste it.