Primary Objectives
1. Prove AI Can Complete Full Development Tasks
Demonstrate that AI is no longer just an assistant — it can deliver complete work:
| What AI Can Now Deliver | Examples |
|---|---|
| Complete features | End-to-end implementation from spec |
| Full test suites | Unit, integration, e2e tests |
| Documentation | READMEs, API docs, inline comments |
| Refactors | Multi-file restructuring |
| Prototypes | Working UI from design specs |
The proof: By end of workshop, devs will have seen AI complete tasks they thought required human coding.
2. Shift the Developer Role
Change what developers do day-to-day:
OLD ROLE: Developer as Code Writer
───────────────────────────────────
Write code → Debug → Write more code → Review → Ship
NEW ROLE: Developer as Orchestrator
───────────────────────────────────
Plan → Define → Validate → (AI executes) → Review → Ship
| Activity | Old Model | New Model |
|---|---|---|
| Planning | 10% of time | 40% of time |
| Defining (specs, acceptance criteria) | 10% of time | 30% of time |
| Writing code | 60% of time | 10% of time |
| Validating (review, test, QA) | 20% of time | 20% of time |
The shift: Developers become architects of outcomes, not typists of code.
3. Mindset Shift: AI Delivers, Human Decides
This is the core mental model for agentic development:
┌─────────────────────────────────────────────────────────────────────────────┐
│ AGENTIC DEVELOPMENT MODEL │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ DEVELOPMENT FLOW │
│ ┌──────────┐ ┌──────────┐ ┌──────────────────┐ ┌──────────────────┐ │
│ │ DEFINE │→ │ DESIGN │→ │ BUILD │→ │ SHIP │ │
│ │ Product │ │ Arch/UI │ │ Develop + Test │ │ Deploy + Run │ │
│ └──────────┘ └──────────┘ └──────────────────┘ └──────────────────┘ │
│ │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ AI DELIVERS (artifacts) │
│ ┌───────────────────────────────────────────────────────────────────┐ │
│ │ • Research & Specs • System Design • Code & Tests │ │
│ │ • Program Plans • Tech Specs • Implementation Plans │ │
│ │ • Epics & Stories • Prototypes • Bugfixes │ │
│ │ • UI/UX Specs • Assets • Pipelines & IaC │ │
│ └───────────────────────────────────────────────────────────────────┘ │
│ │
├─────────────────────────────────────────────────────────────────────────────┤
│ │
│ HUMAN DECIDES (judgment) │
│ ┌───────────────────────────────────────────────────────────────────┐ │
│ │ • Scope & Priority • Trade-offs • Logic & Quality │ │
│ │ • Standards • Experience • Risk & Readiness │ │
│ │ • Usability • Approval • Escalation │ │
│ └───────────────────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Key insight: AI handles the what (deliverables). Humans handle the whether (decisions).
Success Criteria
By the end of this workshop, each developer should:
Believe (Mindset)
- Accept that AI can complete full development tasks, not just assist
- See their role as “orchestrator” rather than “code writer”
- Understand the “AI Delivers / Human Decides” model
Know (Knowledge)
- Explain the difference between autocomplete, copilot, and agentic coding
- Articulate when to use plan mode vs. execution mode
- Describe compound engineering principles
- Know prompt patterns (Specification, Critique, Three Experts, Incremental)
Do (Skills)
- Set up Claude Code with Playwright MCP (and optionally Figma MCP)
- Build an app from a product brief using natural language instructions
- Create and configure a CLAUDE.md for a project
- Use subagents for parallel tasks
- Configure at least one Hook for automation
- Use the Compound Engineering Plugin to plan, build, review, and compound
- Ship a working AI Trust Check MVP to a preview environment
Workshop Details
| Item | Details |
|---|---|
| Format | In-person |
| Duration | 1.5 days (~10.5 hours) |
| Team Size | 7 developers |
| Primary Tool | Claude Code (80% of content) |
| Supporting Tools | GitHub Copilot, OpenAI Codex, Cursor (20% overview) |
| Output | Living framework document (continuously refined) |
| Follow-up | Virtual sessions at 1, 3, and 6 weeks |