Glossary

ABA Frameworks

Terms created and defined by Agent Builder Academy.

AAL - Agent Access Level

How much of your system the agent can access and change. Five levels: AAL1 is a chatbot with no file access. AAL5 is a system that changes itself over time.

AOL - Agent Orchestration Layer

How deeply the AI agent is built into your project. Five layers: from no setup at all (AOL1) to a fully automated system where agents trigger each other (AOL5).

APL - Agent Power Level

How much your agent tools can do and how much control you have over them. Four levels: APL1 is browser chat with no file access. APL4 is multiple agents working together automatically.

Bottleneck

When your control is higher than your agent's access - APL is higher than AAL. Your setup can handle more than it is currently doing. Not a risk, but a sign you are not using the full potential.

Inflection Point

The APL3 / AAL3 boundary where the agent stops giving advice and starts acting directly on your files. Below this point it can only suggest. Above it, it acts.

Mismatch

When your agent can access more than you can control - AAL is higher than APL. This is the main risk in agent setups. The agent can touch things you cannot see or stop.

General Terms

Common terms used in AI development.

Agentic AI

AI that can work toward a goal on its own. It plans what to do, does it, checks the result, and tries again if needed - without you approving every single step.

Assistive AI

AI that waits for your message, replies, and waits again. You are in control of every step. ChatGPT in a browser is assistive. Claude Code running in your project is agentic.

Context

The background knowledge that shapes how a task is interpreted and executed. A human collaborator brings shared context - history, priorities, what "done" means to you. An agent starts with only what you provide and what it can read from the project. Without deliberate context-setting, the agent fills gaps with reasonable assumptions that may not match yours. Context is what you build into your project through CLAUDE.md, /docs, and structured instructions.

Context Window

The agent's working memory - the total amount of text it can read and use at one time. When a session gets long, older parts fall outside the context window and the agent can no longer see them.

Intent

The goal or meaning behind what you asked the agent to do. Intent is what you actually wanted - not just the words you used. What the agent produces is its interpretation of that intent. Past the inflection point, the gap between your intent and the agent's interpretation is no longer visible before the action. It shows up only in the result. Reviewing the diff is how you close the gap.

Prompt

The message or instruction you give an AI. A good prompt is specific about what you want, what format you expect, and what the agent should not do.

Token

The unit AI models use to measure text. One token is roughly one word or part of a word. Models have a limit on how many tokens they can process at once - this is the context window.

Vibe Coding

Building software by describing what you want to an AI agent. The agent writes the code. You guide it by reacting to what it produces - no deep technical knowledge required.