AI Agent Planning Worksheet

Define an AI agent's objective, users, tools, permissions, approval points, limits, failure handling and evaluation criteria in one worksheet you can download and review.

What it helps you do

Agents fail in predictable ways: unclear goals, over-broad permissions and no plan for errors. This worksheet makes those decisions explicit before any code is written.

How to use it

Complete every field. If you cannot fill in the tools or the failure conditions, the agent is not ready to build. The last field asks whether a fixed workflow would do the job instead.

Worksheet

0 of 14 fields completed

01ObjectiveThe single outcome the agent is responsible for, in one sentence.02Users and triggerWho starts the agent and what event or request triggers it.03InputsWhat information the agent receives at the start.04Definition of doneHow the agent, and you, know the task is complete and correct.05ToolsEach tool with its purpose and whether it reads or writes. Keep tools narrow.06Data sourcesSystems and documents the agent can query.07PermissionsThe minimum access each tool needs. Note any access that should be denied.08Human approval pointsActions that need confirmation: payments, deletions, external messages, irreversible changes.09LimitsMaximum steps, time and cost per task.10Failure conditions and fallbackTool errors, missing data, low confidence. What happens in each case, and who is notified.11Memory and stateWhat the agent keeps during a task and between tasks, and for how long.12Evaluation criteria and test casesScenarios to test, including failures, with expected behaviour.13Logging and monitoringWhat is recorded for every run: tool calls, inputs, outputs, cost.14Workflow alternativeCould a fixed sequence of steps achieve the same result? If so, why is an agent needed?

Least privilege

Give each tool the narrowest permission it needs. A tool that can read orders should not also be able to issue refunds unless that is its purpose.

Plan for failure

Decide in advance what happens when a tool errors, data is missing or the agent is unsure. A clear handover to a person is better than a confident wrong action.

Want to build this?

If you have completed the worksheet and want to build the agent, our AI Agent Development team can take it from here.

AI Agent Development Services

Sources & further reading

Building effective agentsAnthropic · 2024 · Engineering articleanthropic.com
A practical guide to building agentsOpenAI · 2025 · Guide (PDF)cdn.openai.com
OWASP Top 10 for LLM ApplicationsOWASP GenAI Security Project · Industry standardgenai.owasp.org

Back to overview AI Resource Hub