AI-First Development

I shipped enterprise software for decades without AI. Then Claude Code changed how I build — forever.

For 20+ years I delivered Fortune-500 systems the long way — every design doc, every line, every test by hand. The past couple of years I’ve gone all-in on AI tooling: research, architecture, diagrams, code, tests, and whole-codebase migrations now run through Claude Code, Gemini CLI, and custom MCP servers, and I review every pass myself. The result on real engagements: roughly 4× delivery acceleration — judged by an engineer who knows exactly what the output should look like. This page is about how I build with AI; building AI into products lives on my AI & Machine Learning page.

~4×
Delivery acceleration observed on real engagements
350+
Codebases analyzed and ported by an agentic pipeline
195K
Lines of code I built AI-first, solo, in six months
20+
Years of engineering behind the AI practice

How I Work AI-First

Six practices, applied on every engagement.

AI-Accelerated Engineering

Research, spikes, and scaffolding in minutes instead of days — so I iterate and test faster, explore more options, and spend my time on the decisions that matter instead of the boilerplate.

Agentic Migration Pipelines

AI agents that analyze and port whole codebases — multi-pass loops that generate documentation, feed it back to the model, and iterate — producing deterministic output that I review on every pass.

Custom MCP Servers

I build Model Context Protocol servers that extend Claude with client-specific tools and context — your repositories, your APIs, your domain knowledge — so the model works inside your world, not a generic one.

Multi-Model Fluency

Claude, Gemini, ChatGPT / Codex-style code models, GitHub Copilot — I pick the model per task and benchmark them against each other on real work, not vendor slide decks.

AI-Assisted Architecture & Docs

Design docs, C4 and Mermaid diagrams, deployment diagrams, and Confluence documentation produced with AI and reviewed by me — documentation that keeps pace with the code instead of trailing it.

Team Enablement

I set up Claude Code for engineers and PMs, teach MCP architecture, and coach teams on prompt engineering — so the acceleration stays with your team after I’m gone.

The Daily Toolchain

The tools I actually build with, every day.

Daily Drivers

Claude Code Claude API (AWS Bedrock & direct) Google Gemini Gemini CLI ChatGPT Codex-style code models GitHub Copilot

Built by Me

Custom MCP Servers Agentic Multi-Pass Pipelines Documentation Feedback Loops Model-vs-Model Benchmarks

The Governance Behind the Speed

Acceleration without oversight is just faster mistakes. Mine comes with rules.

Nothing ships unreviewed

AI output never goes to production without me reviewing it — every migration pass, every generated diagram, every scaffolded service. The model accelerates the work; it doesn’t get the final word.

Deterministic cores where it counts

Where correctness matters, I build deterministic cores with AI language layers on top — the pattern running in Grade My Investments, where ML.NET does the repeatable math and Claude handles the language.

Cost-capped in production

Production AI usage runs under hard spend limits — GMI enforces a monthly Claude cost cap — so the AI-first practice never turns into an open-ended bill.

Want your delivery to move this fast?

I bring the AI-first practice — and the architect who reviews every pass — to your next build or migration. Corp-to-Corp engagements out of Dallas / Ft. Worth.