Advanced Claude Code Techniques for 2026

AI Coding Summit 2026 Workshop | March 3, 2026 | 2 Hours

A hands-on workshop on agentic coding - how to delegate all coding to an AI agent while staying in the driver’s seat. No vibe coding. You’ll learn a structured system where every mistake the agent makes is an opportunity to improve the system itself, making your coding agent more reliable over time.

What You’ll Learn

The PIV Loop: Plan, Implement, Validate

The core workflow that separates productive AI-assisted development from “vibe coding”:

1. Plan - Create context-rich implementation plans that give the agent everything it needs to succeed on the first pass. Not just “build me X” - structured plans with architecture references, code patterns, validation commands, and acceptance criteria.

2. Implement - Delegate all coding to the AI coding assistant. This is a trust-but-verify approach - the agent writes the code, but it’s following your structured plan which includes the full validation strategy so it can check its own work along the way.

3. Validate - The AI validation pyramid. Have the AI coding assistant check as much as possible automatically - linting, type checking, running tests, verifying against acceptance criteria. You step in at the top of the pyramid for code reviews and manual testing. The agent does the heavy lifting; you do the final quality pass.

Golden Nuggets

• Context Reset - Reset the context between planning and implementation. Context is the most precious resource when working with AI coding assistants - don’t pollute your implementation context with all the research and exploration from planning.
• Git Log as Memory - Your commit history is context the agent can use. Structure it well.
• Commandify Everything - Turn your repeatable workflows into reusable commands. This repo includes a full set of commands as a starting point.
• System Evolution Mindset - Every bug from the AI coding assistant isn’t just something to fix - it’s an opportunity to address the root cause in your system. Update your rules, commands, skills, or whatever caused the issue so it never happens again.
• The Sandwich Principle - You own the planning and validation; the agent owns the implementation. That’s how you stay in control while still delegating all the coding.

Workshop Exercise: Poll Builder App

We’ll build a poll/survey builder from scratch - a web application where users can create polls, share them via link, and see real-time results with visualizations.

Why This Use Case?

• Not trivially one-shottable - Has auth, CRUD, relational data (votes), public sharing, and data visualization. Vibe coding this would be a mess.
• Clear PIV loop demonstration - Natural phases that show how planning pays off
• Visually satisfying - You see polls, votes, and charts come to life as we build
• Interactive - We can use the app we build to poll the workshop audience live

What We’ll Build

We’ll create the PRD for the poll builder from scratch, then take that PRD and run a full PIV loop to build out the initial application - authentication, poll creation, voting, and results visualization.

Prerequisites

Before the workshop, make sure you have:

• [ ] Git and GitHub installed and configured
• [ ] An AI coding assistant installed - preferably Claude Code (CLI), but any AI coding tool works
• [ ] Node.js 18+ (or Bun) installed
• [ ] A code editor (VS Code, Cursor, etc.)

Optional (for following along exactly)

• [ ] Claude Pro or Max subscription (for Claude Code access)
• [ ] A database tool (we’ll use SQLite for simplicity, no setup needed)

Repository Structure

ai-coding-summit-workshop-2/
├── .claude/
│   ├── CLAUDE-template.md          # Template for creating project rules
│   ├── commands/
│   │   ├── prime.md                # /prime - Load codebase context
│   │   ├── create-prd.md          # /create-prd - Generate a PRD
│   │   ├── create-rules.md        # /create-rules - Generate CLAUDE.md
│   │   ├── plan-feature.md        # /plan-feature - Create implementation plan
│   │   ├── execute.md             # /execute - Execute a plan step-by-step
│   │   ├── init-project.md        # /init-project - Set up and start locally
│   │   └── commit.md              # /commit - Create atomic git commits
│   └── skills/
│       ├── agent-browser/         # Browser automation for testing
│       │   └── SKILL.md
│       └── e2e-test/              # End-to-end testing orchestration
│           └── SKILL.md
└── README.md

Commands Explained

These are reusable prompts (slash commands) that encode your workflow:

Skills Explained

Skills give the agent specialized capabilities beyond code editing:

The Workshop Flow

1. Overview (~20 min) - High-level overview of agentic coding, the PIV loop, and why structure beats speed
2. PRD Creation (~20 min) - Live demo: creating a PRD for the poll builder from scratch
3. PIV Loop Build (~40 min) - Live PIV loop: plan, implement, and validate to build out the initial application
4. Golden Nuggets & Q&A (~40 min) - Key takeaways, tips, and audience questions

About the Instructor

Cole Medin - AI Engineering educator and content creator. Runs the Dynamous community and creates YouTube content about AI coding and agentic engineering. Previously presented “Advanced Claude Code techniques: agentic engineering with context driven development” at AI Coding Summit 2025.

Resources

• AI Coding Summit 2025 Workshop Repo (last year’s workshop)
• Cole’s YouTube Channel
• Dynamous Community