AI training that ends with running code
I teach engineering teams to build with Claude, OpenAI Codex and Gemini the way I build with them: coding agents let loose on a real repository, agents with tool use and MCP, RAG over real data, evals that catch regressions, and the architecture that holds it all together. No slideware — your stack, your use case, working software by the end.
What your team learns
Building with Claude, Codex & Gemini
Model selection, prompt architecture, structured outputs, tool use, and MCP across all three — plus where each one actually wins. Building blocks your team applies to a real internal use case during the session.
Coding agents in a real repository
Claude Code and OpenAI Codex on your own codebase: scoping a task an agent can actually finish, reviewing its diffs, keeping it inside your conventions and tests, and wiring it into CI — including the tasks where an agent costs more than it saves.
AI agents that survive production
Agent loops, guardrails, evals, and observability. The difference between a demo and a system you can put in front of customers.
Software architecture, taught by doing
Service boundaries, data modeling, and the build-versus-buy calls — reviewed against your actual codebase, not slideware.
Senior habits for the whole team
Code review culture, incident thinking, and how to make technical decisions that hold up. What building at Microsoft, Intel, and Cellebrite since 2010 actually teaches.
Three ways to work together
Team workshop
A focused day (remote or on-site): your stack, your use case, hands-on the whole time. Your team ships a working AI feature by the end of it.
Ongoing mentorship
Weekly or bi-weekly sessions with your engineers: architecture reviews, pairing on the hard parts, and a direct line between sessions.
Embedded training
I join your team for a sprint as a working senior engineer — decisions get made in your codebase, and the reasoning stays with your team.
Why learn from a working CTO
- Shipping since 2010 at Microsoft, Intel, Cellebrite, BigID, and Forcepoint.
- 36 apps in production — the AI patterns taught here run in real products.
- 3 approved US patents; security and privacy are part of the curriculum, not a footnote.
- This site itself is the demo: SSG, structured data, and AI-agent workflows, all built the way I teach it.
Frequently Asked Questions
Who is the AI training for?
Engineering teams that want to build with LLMs properly — startups adding their first AI feature, and established teams that want agents, RAG, and evals done right. Sessions assume working programmers, not beginners.
Which AI models and tools does the training cover?
Claude, OpenAI Codex, and Gemini: their APIs, tool use, agent patterns, MCP, structured outputs, and evaluation. The principles transfer to any model; the exercises run on whichever provider your team already uses.
Does the training cover coding agents like Claude Code and Codex?
Yes, as its own module. We run Claude Code and OpenAI Codex against your repository: scoping tasks an agent can actually finish, reviewing its diffs, keeping it inside your conventions and tests, wiring it into CI — and recognising the tasks where an agent is slower than writing the code yourself.
Is this theory or hands-on?
Hands-on. Every format works on a real use case from your product — the workshop ends with running code in your repository, not a slide deck.
Do you also teach software architecture without the AI part?
Yes. Architecture reviews, system design workshops, and senior-engineering mentorship are available on their own — the AI material is one track, not a requirement.
Remote or on-site?
Both. Remote-first and async-friendly by default, with on-site available where it makes sense.
Tell me your stack and what you want the team to be able to build.