MODELS (WHAT THINKS)
The AI component that actually generates text or code.
What is a model?
An AI that performs a specific task or role.
What is an agent?
The instructions you give to an AI.
What is a prompt?
When humans approve or reject AI output.
What is human-in-the-loop?
AI giving a wrong but confident answer.
What is hallucination?
Different models are often chosen based on speed, cost, or quality.
What is model selection?
An agent focused on gathering information.
What is a research agent?
Telling AI who it should act as.
What is role-based prompting?
Preventing agents from modifying main code directly.
What is worktree or branch isolation?
Repeating research when assumptions change.
What is re-research?
A model optimized for fast, lightweight responses.
What is a “small” or fast model (e.g. Haiku)?
An agent that breaks a goal into steps.
What is a planner agent?
Using the same task but different roles to get different results.
What is role specialization?
Background agents shouldn’t ask users questions.
What is fire-and-forget execution?
When a review rejects work and sends it back.
What is a feedback loop?
A model balanced between reasoning power and cost.
What is a “medium” reasoning model (e.g. Sonnet)?
An agent that checks or critiques the work of others.
What is a reviewer agent?
Asking the same question in a fresh chat to avoid bias.
What is context isolation?
Rules that define what an agent can and cannot do.
What is an agent contract?
more agents can increase cost quickly
What is token burn?
A model used for deep reasoning, planning, or reviews.
What is a “large” or advanced model (e.g. Opus)?
An agent that decides what happens next but doesn’t do the work itself.
What is a controller agent?
When prompts are constrained to allow only specific outputs.
What is a closed or strictly limited prompt?
Separating agent memory by project or scope.
What is scoped memory?
Running multiple independent workflows at once.
What are parallel agent lanes?