// Writing

Notes on shipping AI-built apps to production.

Where agent-generated code breaks, what production-grade actually requires, and how composable, source-visible modules you own hold up. No fluff.

We ran our own admin on our own modules — and it made the product better

We sell composable modules but our own admin used zero of them. We changed that. Using our own product surfaced a real gap in minutes — so we improved the module. Here's exactly what migrated, what didn't, and why.

Read →

Durable, audited background jobs on Cloudflare (without bolting on a queue service)

AI can write the workflow. Production needs durable execution: retries, idempotency, dead-letter, and an audit trail of every run. Compose that on Cloudflare from verified modules you own.

Read →

The white-label SaaS backend agencies rebuild on every client

AI gets each client's SaaS 70% done. The other 30% — auth, idempotent webhooks, multi-tenant isolation, audit — is the same dangerous code every time. Compose it from verified, source-visible modules the client owns.

Read →

I let Claude Code build a Cloudflare app — then broke it the way agents do

84% of devs build with AI; only 29% trust it. I scaffolded a real Cloudflare app, made the classic AI-agent mistake — inlining domain logic — and watched the contract check catch it by name.

Read →

Where AI-built apps break in production

Agents write code that passes tests and reads fine. Then auth, payments, and tenant isolation fail in the places that cost money. Why it happens — and what 'production-ready' actually demands.

Read →

Your agent ships 70% of the app. We make the other 30% safe.

AI gets your app 70% done in an afternoon. The last 30% — auth, Stripe webhooks, email, audit logs, multi-tenant isolation — is where it breaks in production. microservices.sh gives your agent verified, source-visible, Cloudflare-native modules so the dangerous parts are already correct.

Read →