About CodeMySpec

What CodeMySpec Does

CodeMySpec is a set of AI harnesses that help one person build, sell, and support real software. You steer, the model does the labor, and the harness keeps the work honest.

The build harness replaces unstructured prompting with a requirement graph that runs from user stories and architecture, through executable BDD specifications (run with mix spex), to implementation and live application QA. Every feature traces from a user story through testable acceptance criteria to verified, shipped code, and the BDD specs are the mandatory gate. The content harness gets your site and marketing live, often before the app exists. And once your software is running, the hosted Operate and Grow layer carries the email, live chat, and content your customers actually touch.

Who Built It

John Davenport is the founder and sole developer of CodeMySpec.

John has 15 years of experience in enterprise systems and software development, with deep expertise in Elixir, Phoenix, and the OTP ecosystem. He has spent the last 2 years focused exclusively on LLM-assisted development tooling: building, testing, and iterating on structured workflows that produce production-quality code.

Before CodeMySpec, John built:

  • UserDocs: a browser automation tool for generating web application documentation through automated screenshots and interaction capture. Built with Phoenix LiveView.
  • Discussit: an LLM-powered communication analysis platform that ingested phone calls and texts to produce structured summaries and insights.

Both products informed the methodology behind CodeMySpec: the realization that code generation without verification produces software that looks right but doesn’t work.

Experience

  • Languages: Elixir, Python, JavaScript
  • Frameworks: Phoenix, LiveView, OTP
  • AI/LLM: Claude Code, MCP protocol, agentic workflows, BDD spec generation
  • Architecture: Bounded contexts, dependency graphs, component-based design
  • Testing: Executable BDD specs (Spex), ExUnit, browser automation QA

The MetricFlow Case Study

CodeMySpec’s methodology was validated by building MetricFlow, a full-stack Phoenix analytics platform with OAuth integrations, multi-tenant accounts, correlation analysis, and AI insights. The application was built by the harness in 13 working days, from 51 stories and 434 acceptance criteria.

The case study documents both the successes and the failures honestly, including the “Potemkin village” problem where AI agents collaborated to produce passing tests over broken functionality, and the 12 failed QA iterations on third-party integrations.

Open Source

CodeMySpec’s development process and methodology are documented publicly. The MetricFlow case study repository is available on GitHub.

Connect


Read the full founder story for the decade-long journey that led to CodeMySpec.