Citizen Development Pipeline

Background

A North American third-party logistics operator runs warehouse operations across multiple verticals. Their operations team builds the tools they need in Claude. That works to get to the first version. It had not worked for getting those applications onto infrastructure Engineering could trust or maintain.

The gap between what a non-engineer could build and what a production application required was not a skills problem. It was an infrastructure problem. Every operator-built application started from scratch, landed outside the company's standard stack, and created a maintenance burden nobody had signed up for.

Solution

Alphabyte built a two-plugin citizen development pipeline, published in the company's Claude Cowork marketplace. The first run of the pipeline was the warehouse platform rebuild described in the companion case study.

The Spec plugin runs in a chat tab. A non-engineer describes the problem, the users, and the features in plain language. The plugin asks the questions, cross-references the company's existing GitHub organization for prior art, and produces a schema-validated specification document. The project slug is locked from the directory name: deterministic, not user chosen.

The Dev plugin runs in a development tab. The citizen developer says, "build this." The plugin reads the specification and runs a sixteen-step idempotent scaffold: a Next.js application on Node 20, optional authentication with NextAuth and bcrypt, optional Postgres with Prisma and a per-deploy migration hook, optional Resend email, a private GitHub repository under the company's organization, a Railway deploy with Postgres provisioned and the initial migration run, and a public domain. Each step probes filesystem and remote state, skips what is already done, and picks up cleanly from any failure point.

All subsequent development happens in natural-language conversation in the same tab. Dedicated skills handle moments that warrant structure: database schema changes and migrations, Railway deploy and log debugging, third-party API integration grounded in current documentation, the company's design language and tokens, project-memory conventions, and read-only access to the company's curated BigQuery mart layer.

Every application ships on the company's standard stack, into the company's GitHub organization, under the company's Railway workspace, with the same conventions and review patterns Engineering already uses. Shared credentials are auto-fetched cross-projects. Governance is inherited, not configured per application.

Marginal cost of the next operator-built production application is near zero in engineering time. Governance is inherited, not configured per app.