AI agent permissions planner
Define sources, actions, approval, limits and stop controls. Download a draft permissions plan for your AI agent.
An agent needs a clear boundary around what it may read, suggest and execute. This planner turns a general ambition into sources, operations and review points. The result is a draft to discuss with the process owner and developer.
Start with one task, such as preparing a reply to an existing customer enquiry. Specify whether the agent may only draft, or also change a record and send a message. These differences determine the controls to investigate.
Prepare your plan
Your answers stay in this page. Download the draft to keep it; closing or resetting the page clears your input.
Describe access precisely
“Access to the CRM” is too broad for a permissions plan. Which organisation, records and fields are needed? May the agent read only the current enquiry or previous notes too? Which documents are excluded? Describe what happens to access when an employee changes roles.
An instruction to avoid other data is not a substitute for technical authorisation. The application must select allowed records and associate each operation with a valid user or task. The download records the intended rules; a developer must investigate how to enforce them in the systems involved.
Separate a suggestion from an external action
Illustrative example: a support agent may prepare a draft from one product manual and the current enquiry. A member of staff reviews the recipient, content and any attachment. Expanding later to autonomous sending grants a new permission that needs its own tests, limits and recovery arrangements.
Select the highest-impact action in the tool. For payment, deletion or permission changes, the draft recommends explicit review of each irreversible operation. Read-only tasks still need controls on allowed sources. Human approval helps only when the reviewer understands what will happen and the application records the approved operation.
Prepare refusal and stop paths
Include a request outside the allowed task, a source containing misleading instructions and an unavailable reviewer in the evaluation cases. Agree what the agent should return and who takes over. “I do not know” needs a useful next step instead of a dead end.
Download the plan and discuss how to pause operations and revoke access. This tool grants no permissions, connects no accounts and does not assess real security. It produces a concrete set of intended boundaries to test during design, implementation and evaluation.
Understand the decisions behind your plan
AI agents & chatbots ↗AI agent or chatbot: which approach fits your process?
Compare chatbots, fixed workflows and AI agents by task, permissions, reliability and operating effort, with a practical decision example.
Read the guide ↗02 / AI agents & chatbotsWhat does an AI agent cost? Budget for delivery and operation
Build a realistic AI-agent budget covering scope, integrations, evaluation, usage, human review and maintenance, without invented price promises.
Read the guide ↗03 / AI agents & chatbotsWhich permissions should an AI agent receive?
Design bounded AI-agent permissions for reading, proposing and executing, with approval boundaries, practical access checks and recovery examples.
Read the guide ↗04 / AI agents & chatbotsConnecting an AI agent to your CRM: design the data flow
Define an AI-to-CRM integration with dependable customer matching, limited actions, duplicate handling and evidence of actual processing.
Read the guide ↗05 / AI agents & chatbotsTesting an AI agent: turn fluent answers into useful evidence
Build a practical AI-agent test set covering sources, actions, exceptions, handover and repeatable acceptance criteria.
Read the guide ↗06 / AI agents & chatbotsFrom AI agent to employee: design a useful handover
Make AI-agent handover work by recognising boundaries, preserving context, assigning ownership and preventing duplicate follow-up.
Read the guide ↗07 / AI agents & chatbotsBusiness knowledge for AI agents: sources, access and freshness
Prepare AI-agent knowledge sources by defining valid versions, permissions, usable citations, deletion and testable answers.
Read the guide ↗08 / AI agents & chatbotsOperating an AI agent: monitoring, changes and recovery
Plan AI-agent operation with useful alerts, usage limits, source ownership, regression checks, pausing and a workable recovery route.
Read the guide ↗