Fitwell

Fitwell Buckle Co. is a small hardware brand (precision micro-adjust watch buckles) run by two non-technical founders. My favorite test case so far: I built a platform, and a non-technical team stood up their own operational systems on it.

The setup

In about two weeks (May 2026) I built the rails: an admin platform scaffold, data pipelines into their commerce and analytics stack, a test suite, and the governance layer that makes the codebase safe for non-engineers driving Claude Code:

• A single agent playbook (AGENTS.md) loaded into every session, with a context-routing table: before an agent builds anything, a trigger table points it at the decision records it must read first.
• A specs/ tree as durable shared memory: architecture, roles, priorities, strategy, invariants, and a work-plan lifecycle.
• Guardrails as “critical rules”: migration pre-flight checks that block schema changes and explain why, one canonical pricing source, never store payment data, never commit without asking.
• A copy-on-write database branch per person, so nobody can break production or each other.
• A plain-English data assistant: ask questions about the business, get answers with the exact SQL shown, running under a read-only database role that refuses to guess.

What happened

Over the next two months, the two founders out-committed me roughly eight to one, around 750 of 850+ commits. Working through Claude Code on those rails, they shipped: a bookkeeping ledger the business closes the month on, a wholesale portal with a volume-pricing engine, a CRM with sales follow-through, a daily email newsletter with regional sends, production and supplier tracking, and trade-show tooling they’ve used at real shows on two continents. Database migrations included. My ongoing role narrowed to architectural sign-off on new tables and integrations.

Guardrails

Review by a single engineer caps a team’s speed at that engineer’s attention, so the risk was engineered down up front instead: database branches per person, migration pre-flight checks, read-only roles, CI gates on migration collisions, and decisions routed to written records.

Status: In daily production use

The business runs on it: books, sales, marketing, production. The same pattern of handing a non-technical team an AI-operable codebase is part of working with me.