Workshop

AGENTIC
CODING
WORKSHOP

1.5 days. 6 developers. Build with AI, stay in control.

RAB2B · WORKSHOP

AI DELIVERS.
HUMAN DECIDES.

The core mental model for agentic development. AI handles the deliverables—code, tests, docs. Humans handle the decisions—scope, quality, risk.

1.5
Days, In-Person
6
Developers

3 focus areas: Foundations — how agentic coding works. Practice — real workflows on real code. Proof — ship a working component.

RAB2B · WORKSHOP

WORKSHOP
SCHEDULE

DAY 1 9:00 AM – 5:30 PM

  • 9:00 – 12:45 — Foundations: mindset, concepts, frameworks, principles, Claude Code setup
  • 12:45 – 2:45 — Lunch
  • 2:45 – 5:30 — Practice: core workflows, power features, integrations

DAY 2 9:00 AM – 12:45 PM

  • 9:00 – 10:30 — Build from Design: Figma MCP, Playwright MCP
  • 10:45 – 12:00 — Ship It: each dev deploys a component
  • 12:00 – 12:45 — Team Playbook: agree on norms, close out

3 arcs: Foundations (Day 1 AM) → Practice (Day 1 PM) → Proof (Day 2).

RAB2B · WORKSHOP

WHERE YOUR
TIME GOES

OLD MODEL: CODE WRITER

  • Planning — 10% of time
  • Defining (specs, criteria) — 10%
  • Writing code — 60%
  • Validating (review, test) — 20%

NEW MODEL: ORCHESTRATOR

  • Planning — 40% of time
  • Defining (specs, criteria) — 30%
  • Writing code — 10%
  • Validating (review, test) — 20%

The shift: 70% of your time moves to planning & defining. AI writes the code. You orchestrate the outcome.

RAB2B · WORKSHOP

WHAT IS
AN AGENT

An AI system that autonomously takes actions to achieve a goal, using tools, making decisions, and iterating based on feedback.

AUTONOMY

Decides what to do next without explicit instruction.

TOOL USE

Invokes external tools and APIs to accomplish tasks.

GOAL-ORIENTED

Works toward a defined objective, not just responding.

ITERATION

Evaluates results and tries again if needed.

CONTEXT AWARENESS

Reads the environment and adapts behavior.

RAB2B · WORKSHOP

THE AGENT
SPECTRUM

AUTOCOMPLETE

Tab completion

CHAT

Q&A only

COPILOT

Inline suggestions

AGENT MODE

Multi-step tasks

AUTONOMOUS

Background work

↑ Most devs start here

↑ Workshop goal

Less agentic → More agentic. The jump from Copilot to Agent Mode is where real productivity gains happen.

RAB2B · WORKSHOP

THE AGENTIC
LOOP

1. OBSERVE

Read current state

2. THINK

Reason about needs

3. PLAN

Decide on approach

4. ACT

Execute using tools

5. EVALUATE

Check results

The agent repeats this cycle until the goal is met. Each iteration uses feedback from the previous one.

THE CYCLE

Observe → Think → Plan → Act → Evaluate → repeat

RAB2B · WORKSHOP

CONTEXT WINDOWS
& MCP

MODEL CONTEXT SIZES

  • Claude Sonnet/Opus 4 — 200K tokens (~150K lines)
  • GPT-4o / 4 Turbo — 128K tokens (~96K lines)
  • Gemini 2.0 Pro — 2M tokens (~1.5M lines)

Most projects fit in a single context window. Use /compact for long sessions.

MCP: "USB FOR AI"

What: A standard protocol for connecting AI models to external tools.

Why: Standardized connections = portable workflows across tools.

Benefit: Same integration works across Claude Code, Cursor, and more.

RAB2B · WORKSHOP

COMPOUND
ENGINEERING

Each unit of work should make subsequent work easier, not harder.

1

VERIFICATION-FIRST

Always give AI a way to verify its work—tests, types, linting.

INFRASTRUCTURE INVESTMENT

Testing harness + CI/CD = exponential dividends over time.

3

THE COMPOUND LOOP

Plan → Work → Review → Compound → Repeat. Each cycle builds on the last.

RAB2B · WORKSHOP

MATURITY
LADDER

1

MANUAL

No AI. All code hand-written.

2

AI AS REFERENCE

Copy-paste snippets from chat.

3

AI AS ASSISTANT

Inline suggestions, you accept/reject.

4

AI AS PARTNER

Multi-file editing, iterative loops.

5

AI AS AGENT

Compound engineering. Workshop goal.

Most teams are at level 2–3. This workshop takes you to level 5: AI as a full agent with compound engineering principles.

RAB2B · WORKSHOP

COMPOUND AI
SYSTEMS

State-of-the-art AI results come from systems with multiple specialized components, not monolithic models.

RAG

LLM + RETRIEVER + VECTOR DB

Documentation Q&A, context-aware search.

LLM + TOOLS + EXECUTOR

Claude Code: reads, edits, runs, iterates.

MULTI-AGENT

MULTIPLE LLMS + COORDINATOR

Parallel research, writing, and review agents.

Multiple specialized components > monolithic models. MCP enables building these systems easily.

RAB2B · WORKSHOP
Principles

7 PRINCIPLES
FOR AGENTIC WORK

1. Define the Goal, Not the Steps

2. Provide Context, Not Instructions

3. Verify, Don't Trust

4. Iterate in Loops

5. Invest in Infrastructure

6. Use the Best of N Pattern

7. Branch First, Experiment Freely

RAB2B · WORKSHOP

PRINCIPLES
1 – 3

1

DEFINE THE GOAL

DO:

"Fix the bug where users can't log in with special characters."

DON'T:

"Open file X, find function Y, change line Z."

PROVIDE CONTEXT

Share the project's stack, conventions, and constraints—not step-by-step instructions.

CLAUDE.md captures this context so every session starts informed.

3

VERIFY, DON'T TRUST

Always validate agent output: tests, type checking, linting, human review.

This is the compound engineering principle: trust but verify.

RAB2B · WORKSHOP

PRINCIPLES
4 – 7

4

ITERATE IN LOOPS

Agent tries → you review → provide feedback → agent tries again. Faster than specifying everything upfront.

INVEST IN INFRASTRUCTURE

Good CLAUDE.md = better performance. Good tests = safer autonomy. Good CI/CD = faster feedback.

6

BEST OF N

Generate multiple versions, cherry-pick the best. AI creates diverse solutions. You bring judgment to select.

7

BRANCH FIRST

Always create a git branch before major AI changes. Try, fail, retry without risk. Merge only what works.

RAB2B · WORKSHOP

PROMPT
PATTERNS

1. SPECIFICATION

Front-load all context (stack, constraints, patterns), then state the task and requirements.

2. CRITIQUE

Have Claude review its own work for security, edge cases, performance, and accessibility.

3. THREE EXPERTS

Get multiple perspectives: security engineer, performance specialist, UX developer.

4. INCREMENTAL BUILD

Complex features in stages: data model → API → UI → integration → tests.

5. CONSTRAINT

Limit scope to get focused output. "Implement ONLY X. Do not add Y or Z. I'll handle those separately."

RAB2B · WORKSHOP

CLAUDE.MD
& MEMORY

CLAUDE.MD

What: A markdown file in your project root giving Claude persistent context.

Where: /project-root/CLAUDE.md

Why: Claude reads it automatically at session start. Reduces repeated context, ensures consistent style, documents tribal knowledge.

Include: Overview, tech stack, architecture, conventions, common commands.

MEMORY.MD

What: A file where Claude stores learnings that persist across sessions.

Where: .claude/memory.md

Contains:

  • Learned preferences
  • Project discoveries
  • Session notes and patterns

Access with /memory command.

RAB2B · WORKSHOP

POWER
FEATURES

SUBAGENTS

Spawn separate Claude instances for parallel work. Tests while refactoring, research while coding. Give clear, scoped instructions.

HOOKS

Shell commands that trigger at specific events: PreToolUse, PostToolUse, Notification, Stop. Auto-run tests after edits.

SLASH COMMANDS

/help, /clear, /compact, /config, /cost, /doctor, /init, /mcp, /memory, /review, /revert. Plus your own custom commands.

PLAN VS AUTO MODE

Plan: Claude proposes, you approve. Safer for learning.
Auto: Claude executes freely. Faster for routine tasks.

RAB2B · WORKSHOP

SAFETY &
GOVERNANCE

PERMISSION LEVELS

  • Plan Mode — Claude proposes, you approve
  • Auto Mode — Claude executes freely
  • Restricted — Limited tools/directories

SECURITY

  • Never put secrets in prompts
  • Review all generated auth/payment code
  • Use .claudeignore for sensitive paths

DATA PRIVACY

  • Use sandbox/test data in workshops
  • Anonymize sensitive examples
  • Clear sessions after sensitive work

We build a team agreement on Day 2.

RAB2B · WORKSHOP
Lab 1

SETUP &
CONFIGURATION

45
Minutes
  • Install Claude Code and authenticate
  • Create a CLAUDE.md for your test project
  • Complete your first AI-assisted task
  • Explore slash commands and configuration

Open labs/lab1-setup.md to begin.

RAB2B · WORKSHOP

THE CORE
WORKFLOW

Every agentic task follows the same three-step loop.

1

DEFINE

Set the goal, provide context, specify acceptance criteria. The better the definition, the better the output.

VALIDATE

Run the code, check the tests, review the diff. Verify AI output before trusting it.

3

REFINE

Iterate based on what you see. Add improvements, fix edge cases, polish. Repeat until done.

This is faster than trying to specify everything upfront. Define broadly, validate quickly, refine iteratively.

RAB2B · WORKSHOP

BEST OF N
PATTERN

STEP 1: GENERATE

Ask for N different approaches with tradeoffs explained.

STEP 2: EVALUATE

Review all versions. Compare against team patterns and maintenance cost.

STEP 3: SELECT

Pick the best, or combine elements from multiple. Refine the chosen approach.

EXAMPLE PROMPT

"Generate 3 different approaches to this authentication flow. For each, explain the tradeoffs. Then I'll pick one."

AI generates diverse solutions. You bring judgment to select.

RAB2B · WORKSHOP
Lab 2

CORE
WORKFLOWS

75
Minutes
  • Complete a multi-file feature using natural language
  • Apply the "Best of N" pattern to choose an approach
  • Use the branch-first workflow for safe experimentation
  • Experience the Define → Validate → Refine loop

Building a UserProfileCard component. Open labs/lab2-workflows.md.

RAB2B · WORKSHOP

SUBAGENTS:
PARALLEL WORK

HOW IT WORKS

Spawn separate Claude instances for independent tasks. The main agent coordinates, subagents execute in parallel.

"Work on the login page while a subagent writes tests for the API."

BEST PRACTICES

  • Use for truly independent tasks
  • Give clear, scoped instructions
  • Merge results intentionally
  • Avoid sequential dependencies
  • Don't give overlapping tasks

When to use: tests + docs simultaneously, research while coding, refactoring while testing.

RAB2B · WORKSHOP

HOOKS:
AUTOMATION

EVENT
WHEN IT FIRES
EXAMPLE USE
PreToolUse
Before Claude uses a tool
Validate commands
PostToolUse
After Claude uses a tool
Run linter after edits
Notification
When Claude wants to notify
Send Slack message
Stop
When Claude finishes
Run test suite

AUTO-TEST EXAMPLE

PostToolUse hook matching Write|Edit → runs npm test after every file change. Tests run automatically.

RAB2B · WORKSHOP

CUSTOM COMMANDS &
COMPOUND LOOP

CUSTOM SLASH COMMANDS

Create .claude/commands/ with markdown files that define reusable prompts.

  • /review-security — check for vulnerabilities
  • /analyze-failures — parse failure log patterns

Use $ARGUMENTS for dynamic input.

FAILURE LOGGING LOOP

PostToolUse hook logs every failed command to .claude/failures.log.

Then /analyze-failures reads the log, groups by type, identifies patterns, and suggests CLAUDE.md updates.

This is compound engineering: each failure makes future work easier.

RAB2B · WORKSHOP
Lab 3

POWER
FEATURES

60
Minutes
  • Use subagents for parallel work
  • Configure at least one automation hook
  • Create a custom slash command
  • Set up the failure logging compound loop

Open labs/lab3-power.md to begin.

RAB2B · WORKSHOP

DAY 1
RECAP

AM

FOUNDATIONS

Agentic mindset, core concepts, frameworks, 7 principles, Claude Code setup. Lab 1 complete.

CORE WORKFLOWS

Define → Validate → Refine loop, Best of N, UserProfileCard built. Lab 2 complete.

PM

POWER FEATURES

Subagents, hooks, custom commands, compound loop. Lab 3 complete.

Tomorrow: MCP integrations (Figma, Playwright), Ship It challenge, and Team Playbook.

RAB2B · WORKSHOP

SEE YOU
TOMORROW

Day 2 starts at 9:00 AM. Bring your Figma token ready.

RAB2B · WORKSHOP
Day 2

BUILD FROM
DESIGN

Connect AI to external tools. Go from Figma to deployed code.

RAB2B · WORKSHOP

FIGMA MCP:
DESIGN TO CODE

1

CONNECT

Install Figma MCP server. Configure with your Personal Access Token. Verify with /mcp.

EXTRACT

Read Figma files. List components. Extract design tokens: colors, typography, spacing, variants.

3

GENERATE

Claude creates React components matching the design exactly. TypeScript, Tailwind, accessibility built in.

The complete design-to-code pipeline: Figma spec → AI reads design → AI writes component → human validates.

RAB2B · WORKSHOP
Lab 4

FROM BRIEF TO BUILD
THE COMPOUND WAY

75
Minutes
  • Install and use the Compound Engineering Plugin
  • Rebuild AI Trust Check with structured workflows
  • Compare Round 1 (unstructured) vs Round 2 (compound)
  • Deploy a working app to a preview URL

Open labs/lab4-compound.md to begin.

RAB2B · WORKSHOP

PLAYWRIGHT MCP:
BROWSER AUTOMATION

CAPABILITIES

  • Navigate to any URL via natural language
  • Fill forms, click buttons, read content
  • Take screenshots of what Claude sees
  • Generate E2E test files from browser sessions

TESTING ANGLE

  • Test your app by describing behavior in English
  • Claude writes Playwright tests from what it observes
  • Compare rendered UI against Figma specs

Compound AI: Figma MCP + Playwright MCP = visual regression testing.

RAB2B · WORKSHOP
Lab 5

PLAYWRIGHT
MCP

45
Minutes
  • Connect Claude Code to Playwright via MCP
  • Automate a browser workflow using natural language
  • Create a simple E2E test
  • Experience AI-driven browser automation

Open labs/lab5-playwright.md to begin.

RAB2B · WORKSHOP
15
Minute Break
RAB2B · WORKSHOP

THE WORKFLOW
END TO END

EXTRACT

Figma → Spec

IMPLEMENT

AI writes code

VALIDATE

Tests + review

REFINE

Iterate + polish

DEPLOY

Ship to preview

STEP
HUMAN
AI
Extract
Points to Figma file
Reads design, extracts spec
Implement
Reviews approach
Writes code, tests, docs
Validate
Checks output, runs tests
Self-reviews, fixes issues
Refine
Gives feedback
Iterates on changes
Deploy
Approves and ships
Prepares build
RAB2B · WORKSHOP
Lab 6

FIGMA
MCP (BONUS)

45
Minutes
  • Connect Claude Code to Figma via MCP
  • Extract design tokens from a Figma file
  • Generate component code from a design spec
  • Experience the design-to-code workflow

Open labs/lab6-figma.md to begin.

RAB2B · WORKSHOP

SHOW
& TELL

Share your deployed component URL with the group.

MATEO MATEUS

URL: _________________

TATIANA TERRONT

URL: _________________

ALEJANDRO CAMACHO

URL: _________________

JOHAN VALBUENA

URL: _________________

NICOLAS SUAREZ

URL: _________________

MIGUEL ARBELAEZ

URL: _________________

RAB2B · WORKSHOP

TEAM
AGREEMENT

SAFETY & GOVERNANCE

  • What requires human approval?
  • Which directories/files are off-limits?
  • How do we handle secrets?
  • AI code review process?

WORKING AGREEMENTS

  • When to use Plan Mode vs Auto Mode
  • How we share learnings across the team
  • Shared definitions and vocabulary

QUALITY GATES

  • Tests pass before merging
  • Human review complete
  • Documentation updated
  • Security check done

TOOLS & PLUGINS

  • Claude Code as primary tool
  • MCP integrations we'll use
  • Custom commands & hooks
  • Review schedule: weekly, monthly, quarterly
RAB2B · WORKSHOP

MAKING
IT STICK

1

WEEK 1

Sync on what landed and what's blocked. Troubleshooting session with the team. Fix early friction.

WEEK 3

Working session on a topic the team picks. Flag what's working and what needs adjusting.

6

WEEK 6

Team owns the playbook. Final review, close out gaps, full handoff. Agentic workflows are routine.

The workshop gives the team the tools. The follow-up sessions make them stick.

RAB2B · WORKSHOP

AI DELIVERS.
HUMAN DECIDES.

You shipped it. Now make it stick.

RAB2B · WORKSHOP
1 / 47