Two complementary frameworks for understanding and applying agentic development.
Compound Engineering
Source: every.to/guides/compound-engineering
Core Principle: Each unit of work should make subsequent work easier, not harder.
The Compound Loop:
Key Concepts:
| Concept | Description |
|---|---|
| Verification-first | Always give AI a way to verify its work (tests, types, linting) |
| Infrastructure investment | Testing harness + CI/CD = exponential dividends |
| Mindset shift | Code writer → System orchestrator |
| Time allocation | 80% planning + review, 20% code generation |
The Maturity Ladder:
- Manual development (no AI)
- AI as reference (copy-paste snippets)
- AI as assistant (inline suggestions)
- AI as partner (multi-file editing)
- AI as agent (compound engineering) ← Workshop goal
Compound Engineering Plugin
Every.to has released an official Claude Code plugin that implements these principles:
Install:
/plugin marketplace add https://github.com/EveryInc/compound-engineering-plugin /plugin install compound-engineering
Commands:
| Command | Purpose |
|---|---|
/workflows:plan | Turn feature ideas into detailed implementation plans |
/workflows:work | Execute plans with worktrees and task tracking |
/workflows:review | Multi-agent code review before merging |
/workflows:compound | Document learnings to make future work easier |
Why it matters: This plugin codifies the compound engineering workflow — plans inform future plans, reviews catch more issues, patterns get documented.
Similar Plugins Worth Exploring
These plugins are optional and not used in the workshop labs. Explore them after the workshop as you develop your own workflow.
| Plugin | Purpose | Install |
|---|---|---|
| Ralph Wiggum | Autonomous iteration loops — Claude works repeatedly until completion | /plugin install ralph-wiggum |
| Context7 | Live documentation lookup — reduces hallucinations with current API docs | /plugin install context7 |
| Feature Development | Structured workflow: explore → design → implement → review | /plugin install feature-development |
| Security Guidance | Real-time security scanning during code generation | /plugin install security-guidance |
Compound AI Systems
Source: Berkeley AI Research
Core Principle: State-of-the-art AI results come from systems with multiple components, not monolithic models.
Architecture:
- Multiple specialized components working together
- LLM + retriever + tools + planners + validators
- Each component optimized for specific function
Key Patterns:
| Pattern | Components | Example |
|---|---|---|
| RAG | LLM + retriever + vector DB | Documentation Q&A |
| Agent | LLM + tools + executor | Claude Code |
| Multi-Agent | Multiple LLMs + coordinator | Research + Writing agents |
| Hybrid | Neural + symbolic | Code generation + linting |
Why It Matters:
- Client prototypes often need compound systems (not just a single prompt)
- MCP enables building these systems easily
- Understanding this = better architecture decisions
How the Frameworks Apply
| Framework | Answers | Workshop Application |
|---|---|---|
| Compound Engineering | How do we work with AI? | Planning, verification, iteration patterns |
| Compound AI Systems | What do we build with AI? | Multi-component architectures, MCP pipelines |