John Davenport

Founder, CodeMySpec

Full-stack engineer with 20 years of experience spanning enterprise systems, manufacturing IT, fintech, and AI-assisted development tooling. I build tools that help developers ship production software with AI — not by replacing architects, but by giving AI agents the structured workflows they need to write code that actually works.

CodeMySpec is my current project: an AI development harness for Elixir/Phoenix applications that enforces the full loop from user stories through architecture, BDD specs, implementation, and QA. I built three production applications with it — MetricFlow (40 commits, 13 days, zero human-written code), Fuellytics (a fuel card fraud detection platform with 5-validator pipeline, Claude Vision OCR, and Stripe Issuing integration), and Market My Spec (a small MCP server that exposes a marketing-strategy skill to Claude Code, shipped across one weekend with zero prompts and zero lines of code) — to prove the methodology works before asking anyone else to use it.

Career Path

I started in hardware design at GE Intelligent Platforms (2005-2010), designing embedded computing solutions — VME and CompactPCI single board computers for military, aerospace, and industrial customers. That gave me a foundation in systems thinking that's shaped everything since.

From there I moved into manufacturing IT at Enova Premier, managing production systems across 4 automotive plants for a tier-one supplier assembling wheels and tires for Toyota, GM, and Hyundai. I built an industrial IoT platform combining OPC UA, real-time PLC data, and GraphQL APIs into operator-facing Vue.js applications — and hired and trained the first software development team the company ever had.

I spent three years at FunnelCloud as Head of Implementations, leading technical operations for an MES vendor serving 50-60 automotive plants. Then three years at Openforce as a software engineer on IC Insure — an insurance marketplace built on an Elixir/Phoenix monolith with CQRS, Event Sourcing (Commanded), and GraphQL (Absinthe), serving clients including American Express and Hudson Insurance Group.

Along the way I founded several products that taught me more than they earned: UserDocs (documentation automation, three major architectural rewrites, secured a pilot with Chef), Discussit (conversation intelligence with LLM-powered topic analysis), and Desert First Cleaning (which actually did work — $252K+ revenue, 150% YoY growth).

What I've Built

  • CodeMySpec — AI development harness for Phoenix. 368 modules, 46 bounded contexts, compile-time dependency enforcement. MCP servers for product management and architecture, BDD spec generation, session orchestration, and automated QA. Ships as a Claude Code extension.
  • Fuellytics — Fleet fuel card fraud detection platform. 5-validator pipeline, Claude Vision OCR, Stripe Issuing/Treasury/Connect, Twilio SMS. Built with CodeMySpec in 55 commits, zero human-written code.
  • MetricFlow — Cross-platform ad analytics correlating Google Ads, Facebook Ads, and GA4 with QuickBooks revenue. Pearson correlation engine, Claude-powered insights. 40 commits, 13 days, honest assessment.
  • Market My Spec — Small MCP server that exposes a marketing-strategy skill to Claude Code. Multi-tenant accounts with agency tier, magic-link auth, OAuth scaffolding, eight-step strategy walkthrough delivered over MCP. 17 commits across 4 days, zero prompts, zero lines of code.
  • IC Insure — Insurance marketplace for independent contractors. Elixir/Phoenix, CQRS/Event Sourcing, 11 bounded contexts, 20+ supervised processes. One of two engineers.
  • UserDocs — Browser automation for documentation. Phoenix LiveView, Chrome DevTools Protocol, multi-tenant architecture. Three major rewrites.
  • Discussit — Conversation intelligence platform with Zoom/OpenPhone ingestion, transcription, and LLM-powered topic clustering.

Technical Focus

Languages & Frameworks

Elixir, Phoenix, LiveView, OTP, Python, TypeScript, C#, Blazor, Vue.js, React

AI & LLM Engineering

Claude API, MCP servers, agentic workflows, BDD spec generation, embedding pipelines, RAG

Architecture

CQRS, Event Sourcing, DDD, bounded contexts, dependency graphs, multi-tenant systems

Data & Infrastructure

PostgreSQL, Snowflake, SQLite, Docker, Fly.io, GitHub Actions, OPC UA, industrial IoT

Connect

Published Articles

Naive BDD: The Tests Ran, the Tests Passed, the App Didn't Work May 03, 2026
Prompt and Pray: How I Started With AI Agents, and Why It Broke May 03, 2026
Spec-Driven Development: Finally Useful, Still Not Executable May 03, 2026
BDD Attention Thesis: A Five-Step Workflow for AI Coding That Doesn't Drift May 03, 2026
Three Amigos: The Gate That Was Missing May 03, 2026
Write It Down: The First AI Coding Workflow That Actually Worked May 03, 2026
Dedicated Memory Stores: Most Marketing, Messiest Tradeoffs Apr 25, 2026
Graph and Structured Memory: Most Ambitious, Most Vaporware Apr 25, 2026
It's Not Memory If You Wrote It: The CLAUDE.md Confusion Apr 25, 2026
Repo-Native Memory: The Boring Answer That Wins Apr 25, 2026
Retrieval and RAG: The Category Everyone Reaches for First Apr 25, 2026
Transcript-Derived Memory: The Category Nobody's Writing About Apr 25, 2026
Build In Public: The Three Amigos Problem Apr 20, 2026
Why My Harness Produced Incomplete Apps (and What I Changed) Apr 19, 2026
What Is a Spec? The Most Overloaded Word in Software Apr 17, 2026
Opus 4.7 Migration Guide: What Breaks, What's Better, What to Watch Apr 16, 2026
The Skill Trajectory for Working with AI Agents Apr 11, 2026
How I Built a Local Embedding Pipeline in Elixir That Searches My Own Docs Apr 10, 2026
How the CodeMySpec Feedback Widget Captures Screenshots in LiveView Apr 10, 2026
What Is the Client? (It's More Powerful Than You Think) Apr 10, 2026
What Is It and Where Does It Belong? Apr 10, 2026
What Is Code, Actually? Apr 09, 2026
What Is a Server and Why Are You Paying For One? Apr 09, 2026
Your App Has Two Halves and You Need to Know Which Is Which Apr 09, 2026
ADRs Are the Best Thing You Can Give Your AI Agent Apr 08, 2026
Anthropic Just Made the Agent Harness a Product Apr 08, 2026
Claude Mythos: What We Know, What We Don't, and Why the Harness Still Matters Apr 08, 2026
Cursor 3 Isn't an IDE Anymore. It's an Agent Switchboard. Apr 08, 2026
What Is Progressive Disclosure and Why It Matters for AI Agents Apr 08, 2026
Architecture for AI Agents: Which Patterns Actually Work? Apr 02, 2026
The Implementation Phase: AI Writes the Code, But Who's Actually Driving? Apr 02, 2026
Maintenance: Where Agents Actually Earn Their Keep Apr 02, 2026
The Verification Gap: Why Agents Ship Broken Code and What to Do About It Apr 02, 2026
Bad Requirements Are Why Your AI Agent Writes Bad Code Apr 02, 2026
The Agentic Software Development Process Apr 02, 2026
Your AI Agent Is Only as Good as Your Spec Apr 02, 2026
Testing AI-Generated Code: The Self-Confirming Loop and How to Break It Apr 02, 2026
Teaching AI Agents to Deploy: Knowledge Files vs. Direct Access Apr 01, 2026
How I Do Marketing with Claude Code and MCP Tools Apr 01, 2026
The Five Layers of an Agentic Coding System Mar 31, 2026
The Agent Layer: How AI Coding Tools Actually Work Mar 31, 2026
The Environment Layer: Where AI Code Actually Runs Mar 31, 2026
The Harness Layer: Why the Wrapper Matters More Than the Model Mar 31, 2026
The Model Layer: What Your AI Coding Tool Actually Is (and Isn't) Mar 31, 2026
The Orchestration Layer: Coordinating Multiple Agents Mar 31, 2026
GitHub Copilot in 2026: Features, Pricing, Benchmarks, and Community Sentiment Mar 30, 2026
Aider in 2026: Features, Pricing, Benchmarks, and Community Sentiment Mar 29, 2026
Gemini CLI in 2026: Features, Pricing, Benchmarks, and Community Sentiment Mar 28, 2026
Codex CLI in 2026: Features, Pricing, Benchmarks, and Community Sentiment Mar 27, 2026
Cursor in 2026: Features, Pricing, Benchmarks, and Community Sentiment Mar 26, 2026
Writing Applications for LLMs Mar 26, 2026
Claude Code in 2026: Features, Pricing, Benchmarks, and Community Sentiment Mar 25, 2026
Open Source vs Vendor-Locked AI Coding Tools: The Tradeoffs That Matter Mar 24, 2026
What Happened to Supermaven, Aide, and Void: AI Coding Tools That Didn't Make It Mar 23, 2026
Building a Markdown API for LLM Collaboration Mar 23, 2026
CodeMySpec Specs vs Kiro EARS: Two Approaches to Spec-Driven AI Development Mar 22, 2026
MCP: The Protocol Connecting AI Coding Tools Mar 21, 2026
Best Free and Open-Source AI Coding Tools in 2026 Mar 20, 2026
AI IDEs Compared in 2026: Cursor vs Windsurf vs Zed vs Kiro Mar 19, 2026
Claude Code Skills: Writing Apps for Agents Mar 19, 2026
The Rise of CLI Coding Agents: Why Terminal-Native AI is Having a Moment Mar 18, 2026
The Best CLI Coding Agents in 2026: Claude Code vs Codex vs Gemini CLI vs Aider vs OpenCode vs Goose Mar 17, 2026
The Five Levels of AI-Assisted Development Mar 15, 2026
Agentic QA Mar 10, 2026
BDD Specs for AI-Generated Code Mar 10, 2026
How CodeMySpec Built and Verified a Fuel Card App in 5 Days Mar 10, 2026
The Part Nobody Talks About: Verifying AI-Generated Code Mar 10, 2026
Remote Permission Approval: Building Trust Boundaries for Autonomous AI Agents Mar 08, 2026
My first serious coding workflow with AI Nov 13, 2025
How to write design documents that keep AI from going off the rails Nov 13, 2025
How to design architecture that keeps AI on track Nov 07, 2025
Why Phoenix Contexts Are Perfect for LLM-Based Code Generation Nov 07, 2025
How to manage user stories to get the most out of LLM's Oct 29, 2025
Code Generation is About Control, Not Prompts Oct 28, 2025
Design-Driven Code Generation, The Missing Layer Between Specs and Code Oct 28, 2025