Agent Orchestration Layers
Context and Intent - both grow with each AOL layer. More context means clearer intent. Clearer intent means better output.
Overview
AOL has 5 layers. Each layer adds more context and more intent. The jump from AOL 2 to AOL 3 is where output quality makes a significant leap - the agent moves from basic orientation to understanding your actual standards and patterns.
The Framework
AOL measures two things: the context your agent has about your project, and the intent it can act on as a result. Two builders using the same tool at the same APL can have very different AOL levels depending on how much structure they have built. The examples below use Claude Code. Other tools apply the same principle under different names.
No project files. No memory. No rules.
- General training knowledge
- Whatever you type in the prompt
- Any knowledge of your project
- Your rules or standards
- Memory between sessions
Create a CLAUDE.md file at your project root.
+ CLAUDE.md. The agent knows where it is and what the rules are.
- Project identity and purpose
- Build and test commands
- Core rules to follow
- Your design patterns
- Architecture decisions
- Project-specific standards
Add a structured /docs folder with standards and patterns.
+ /docs + MEMORY.md. The agent knows your standards and accumulates project memory across sessions.
- Design standards and token references
- Architecture decisions and reasoning
- Anti-patterns to avoid
- Business logic and domain rules
- Accumulated project memory (MEMORY.md)
- Custom procedures for repeatable tasks
- Automated quality checks
Add skill files in .claude/commands/ for repeatable tasks.
+ Skills. The agent picks up the right tool and executes it precisely.
- Custom procedures for repeatable tasks
- Multi-step task execution
- Consistent, predictable output per task
- Automatic triggers
- Self-initiated quality checks
Configure hooks in .claude/settings.json to trigger skills automatically.
+ Hooks. The agent checks its own work and corrects course without being asked.
- Automatic triggers on events
- Self-initiated quality checks
- Proactive error detection
- Full Autonomy Zone coverage
- Skills must be solid before hooks add value
- Human in the loop shifts to review, not direction
The full orchestration stack. Map, Blueprint, Tools, and Automation in place.
AOL is independent of APL
A builder at APL 2 (in-editor assistant) can have a high AOL if they have built strong context and clear intent into their project. A builder at APL 3 (full file access) can have a low AOL if they have never created any project structure for the agent.
AOL does not increase risk. It increases precision. The AI Setup Snapshot measures your AOL alongside your APL and AAL and shows you where to invest next.
Find your AOL, APL, and AAL in 3 minutes.
Take the AI Setup Snapshot