Give agents boundaries
Separate plans, proposals, approvals, and execution. Give agents narrow tools, external authorization checks, and a tested stop control.
Read the guideDAPP CMS LAB / CATEGORY / 2 GUIDES
AI & Agents.
A model can suggest a draft; a tool can change a system. These guides examine that boundary through reviewed editorial workflows and deliberately scoped agent capabilities.
Start with a narrow task that produces an inspectable result. The drafting guide addresses source material, output validation, and human editorial review. The agent guide goes further into concrete proposals, authorization, tool scope, and stopping work. Keep these designs distinct: a useful draft-only assistant does not need permission to publish, modify configuration, or initiate a transaction. Re-evaluate the architecture whenever a new action capability is introduced.
Explore the topic guideA QUESTION TO KEEP IN VIEW
Which control outside the model decides whether the proposed action is authorized?
AI drafts. People decide.
Scope a useful AI drafting task, preserve source evidence, validate the output, and keep editorial authority with the reviewer.
Read the guide