The Mental Model
Replace the autocomplete model with the one that explains both the wins and the failures. Install the supervision stance, the briefing discipline, and the decomposition habit that the rest of the course builds on. Ships seven working assets including the Jagged-Frontier Map, the Brief Template, and the Trust-Calibration Policy.
Replace the autocomplete mental model with the one that explains both the wins and the failures you actually see
Diagnose why most 'the AI is bad at this' moments are briefing failures, not capability failures
Decompose coding work at the seams that keep each unit inside the model's reliable zone
Adopt the accountability model that governs every AI-assisted line you ship: you are the author, the model is the contributor
Claude Code Mastery
Project memory, permission models, mission scoping, and the plan-reading skill that separates supervising an agent from being surprised by one. Ships the CLAUDE.md starter kit and the mission-scoping playbook.
Build a CLAUDE.md that loads your project's load-bearing context into every agent session automatically
Configure what the agent may do autonomously versus what requires your approval, deliberately rather than by default
Scope an agentic mission tightly enough that the agent's context stays bounded and its work stays verifiable
Use plan-first workflows to catch flawed approaches before any code is written, when correction is cheapest
Quality Engineering
Test-first with AI, review pipelines, the regression ritual, and debugging with AI. Ships the Test-First Loop card, the Review Pipeline checklist, the Fresh-Context AI-Review prompt, the Release-Day Runbook, the AI Debugging Practice, and the Debugging Verification Sequence.
Make the test the definition of done so the agent has a ground-truth target to generate against and self-correct toward
Build a layered review pipeline where machines catch what machines catch best, so your attention goes where only judgment works
Run the release-day ritual that turns every model and tool update from a gamble into a one-hour experiment
Use AI to accelerate diagnosis — generating hypotheses, reading unfamiliar errors, narrowing the search — without outsourcing the judgment of what is actually wrong
The API and Agents
Turn Claude into a production service: the API contract, the agent loop, MCP for governed internal capabilities, and the production-reality engineering that makes a shipped AI feature viable at scale. Ships the Production Call Template, Tool Design Card, Execution-Gap Checklist, Build-vs-Connect Guide, Governed MCP Server Checklist, Production-Reality Engineering Guide, and Architectural Levers.
Move from chat to the API as the shift from conversation to a programmable, versioned contract you own
Understand the agent loop mechanically — the model proposes a tool call, your code executes it, the result returns, repeat
Understand MCP as the standard that lets tools and agents interoperate without bespoke integration for each pairing
Treat a shipped AI feature as a production system with cost, latency, and reliability characteristics, not a demo that happens to call a model
The Evaluation Discipline
Eval suites, grader selection, ship-bar discipline, and injection security for the AI features your team ships. Ships the Feature Eval Suite Guide, Ship-Bar Template, Injection Threat-Model Worksheet, and Adversarial Eval Set Guide.
Build an eval suite for an AI feature your team ships, treating its behavior as a product quality system rather than a vibe
Understand prompt injection as the defining security problem of LLM systems: the model cannot reliably separate instructions from data
The Team Playbook
Scale from your personal workflow to your team's standard: the team CLAUDE.md system, the AI-Assisted SDLC Playbook, onboarding new developers into the practice without diluting it, and measuring AI's impact honestly. Ships 8 assets: Team CLAUDE.md System Guide, Promotion-and-Pruning Pipeline, AI-Assisted SDLC Playbook Template, Adoption-and-Metrics Plan, AI-Assisted Onboarding Guide, Onboarding-as-Improvement Practice, AI-Impact Measurement Guide, and Honest-Measurement Discipline.
Scale project memory from a personal file into a governed, shared standard your whole team's agents run on
Assemble the full course into a team standard: how AI-assisted work is briefed, supervised, verified, secured, and shipped
Bring new developers into the team's AI-assisted practice quickly, so they inherit the standard rather than reinventing or undermining it
Measure whether AI-assisted development is actually helping your team, using signals that resist the easy illusions