Generative Agent Optimization (GAO)
Structure your knowledge so AI agents discover, install, and act on it.
Definition
Generative Agent Optimization is the discipline of structuring your organization's knowledge so AI agents can discover it, install it into their context, and act on it deterministically. It's the third layer after SEO (search) and GEO (answer engines).
The Three-Layer Discipline
SEO — Search Engine Optimization
Humans search, read, click
↓
GEO — Generative Engine Optimization
LLMs cite your answer in responses
↓
GAO — Generative Agent Optimization
Agents install your knowledge before acting
How Agents Discover Knowledge
Agents install knowledge through three surfaces:
- llms.txt — Machine-readable site description
- MCP servers — Tools that expose structured knowledge (CKGs, APIs, databases)
- Compact Knowledge Graphs — Pre-compiled domain ontologies delivered via MCP
Why Agents Need Structure (vs. Humans)
❌ Without GAO — Agent acts blind
Engineer writes: "Refactor UserService.ts" — Agent doesn't know 23 modules import it. RAG returns similar code. Agent refactors. Breaks production. Blast radius was invisible.
✓ With GAO — Agent compiles knowledge first
Same request → Agent compiles the knowledge graph of your codebase. trace_downstream("UserService") returns exact 23 dependent modules. Agent sees the blast radius before writing a line. Edit is safe.
The GAO Playbook
Four steps to make your knowledge agent-ready:
- Audit your knowledge. What does your domain depend on? What breaks if X changes?
- Compile it. Build a CKG (Compact Knowledge Graph) from your domain data. Use the CKG Compiler to validate structure.
- Publish it. Expose via llms.txt, MCP server, or direct API so agents can query it.
- Install it. Agents discover and load your knowledge before they act.
Who Needs GAO?
Any organization where AI agents make decisions that affect production:
- Engineering teams (agents refactoring code, shipping features)
- Product teams (agents analyzing customer data, making trade-offs)
- Operations (agents managing infrastructure, responding to alerts)
- Finance & legal (agents reviewing contracts, identifying risk)
The Measurable Outcome
Organizations that implement GAO see:
- ✓ Zero blind edits — agents know the blast radius before acting
- ✓ 42× fewer tokens — compiled knowledge costs 274 tokens, not 17,900
- ✓ 0% hallucination rate — every relationship is explicit, cited, typed
- ✓ Audit trail — every agent decision is traceable to a typed edge in your graph
Get Started
Install ckg-mcp and compile your first domain:
$ pip install ckg-mcp
$ ckg-mcp compile your-domain.csv
Learn more: What is a Compact Knowledge Graph?