Build AI Agents That Ship

Learn AI-assisted development the way you'll actually do it — inside one real codebase seeded with twelve deliberate bugs. Every lab runs three ways (Claude Code, Copilot, or a free tier), every fix has to be verified, and the final track has you measure your own before-and-after honestly.

The same course and the same twelve deliberate bugs, rebuilt on a FastAPI + SQLAlchemy backend you run with uv. Every lab still runs three ways (Claude Code, Copilot, or a free tier), and every fix was re-measured on the Python code — including the places where Python behaves differently.

Thinking about the AI path and not sure where to begin? Four starter PDFs, in order, on one study page. Stop 1 is the map: The Autonomy Line, where AI work is heading. Stop 2 covers the coding assistants you'll use on day one, stop 3 what an agent really is, and stop 4 how to build one properly, with a visual guide of 11 diagrams, an eval rubric and 30 interview flash cards, also as a PDF. Each stop has key takeaways, a glossary and self-check questions, with the full PDF one click away.

Full agent lifecycle — design, build, protect, observe, deploy. 9 tracks from LLM fundamentals through multi-agent production systems. Includes certification prep for Claude Certified Architect.

Deep-dive into Claude Code — API essentials, prompt engineering, CLAUDE.md, permissions, subagents, hooks, tool use, MCP, and the Agent SDK. Fast path for engineers already comfortable with LLMs.

Build, secure, and orchestrate MCP servers from scratch. Covers resources, prompts, HTTP/SSE transport, multi-server orchestration, and production patterns for Claude Code and IDE integrations.

Stop prompting and start specifying — a field guide to specs Claude can build against. Three scenarios (feature, defect, migration) worked end-to-end on a healthcare prior-auth platform, with a runnable Spring Boot companion repo and four hands-on labs.

The same discipline, tool-assisted. OpenSpec keeps capability specs as living documentation and every change as a reviewable proposal that is archived back into them — across Claude Code, Cursor, Copilot and 30+ agents. Nine phases from setup through governance, in Python and Java.

The AI-Driven Development Lifecycle run with AI-DLC Workflows on Claude Code. Start by setting up a workspace, with every command and every generated file explained; then the concepts (phases, stages, gates, the audit record), a one-page cheat sheet, and a 17-step hands-on run that takes one real requirement from a Slack message to a merged pull request — on a Spring Boot repo, or the same run on a Python (FastAPI) repo. Windows and macOS commands, real code and real output. Plus a variant that starts from a Rally feature and a SharePoint spec.

Master the LLM context window end to end — design, assembly, retrieval, compression, optimization, and governance of everything you feed a model. The skill under every reliable agent — and under every good session with a coding agent: every module carries a Dev Lens naming its form in your IDE, from CLAUDE.md as a system prompt to what /compact quietly drops.

From RAG limits to self-updating codebase brains — structural graphs (tree-sitter, Graphify), Google's Open Knowledge Format, MCP serving, and honest multi-arm token benchmarks measured on real repositories.

Master the discipline on an app that shows its tokens — MedFlow Copilot's Token Lens itemizes every AI call by layer, cost, and correctness. Ten strategies, thirty labs, and one law: savings only count when the answer is right.

Beginner-friendly track for Google's AI pair programmer in VS Code and JetBrains — completion, chat, GEMINI.md context, agent mode, MCP, Gemini CLI, GitHub PR reviews, Cloud Run deployment, and enterprise controls.

End-to-end track for Gemini CLI — prompting, GEMINI.md, skill files, plan mode, requirements, design, implementation, testing, deployment, MCP servers, extensions, GitHub Actions, and Google Cloud MCP.

Build production agents using open-source models. Covers the full agent stack — RAG, memory, multi-agent systems, guardrails, observability, deployment, and cost optimization — without vendor lock-in.

Build the same AI agent three ways — in the Anthropic (Claude), Google (Gemini), and OpenAI SDKs, side by side. Tool use, structured output, memory, RAG, multi-agent, and production, each mapped across all three so switching providers is a config change, not a rewrite.

Build a prior-auth portal's support agent with LangChain and LangGraph: it answers from live data when it can, raises a well-formed ServiceNow ticket when it can't, and never pretends otherwise. Graphs, interrupts, token economics, evals with a rubric judge, and a night-shift capstone, all offline on a one-click simulator.

Side-by-side analysis of Claude Code vs Gemini CLI vs GitHub Copilot. Covers setup, prompting, code generation, file ops, agentic workflows, MCP, CI/CD, cost, and how to pick the right stack.

The "Day 2" discipline for LLM systems already in production — evaluation, deployment, monitoring, incident response, drift, and governance. Keep agents reliable, observable, and accountable after launch.

Build the internal AI platform layer every team relies on — model gateway, prompt registry, guardrails, evaluation harness, cost controls, and compliance. Turn scattered AI experiments into shared infrastructure.

AI adoption for senior delivery professionals, on one synthetic prior-authorisation estate with real defects planted in it. A shared core, then a pack for your discipline — QA, performance, SQL, architecture, or engineering. Every case ends in an artifact you file, every lab runs dual-path (Claude Code or Copilot), and you measure your own before-and-after instead of being handed a number.

Take one workflow from an assistant suggesting a line to an agent running whole cases, and build the controls that earn each step. Nine modules on prior-authorisation intake, with nothing to install: every lab runs on evidence shipped with the course, including the agent output that is plausible and wrong. Each module ends in something you keep — a context file, a review checklist, an autonomy boundary, a golden set, an agent card.

The argument the practitioner course is built on, written for the people who fund the work: autonomy is earned, one station at a time. Enterprise AI adoption drawn as a transit line, from coding assistants to orchestrated agents, with the platform, guardrails, people and value work each station needs. A six-question assessment places your organisation on the line, and the route card prints on one page.