Lesson 5 of 5 · about 5 min
Safe first workflows, permissions and cost
When you finish, you can plan a first low-risk agent task with narrow permissions, a spending limit and a review routine.
Stuck? Ask the AI tutorPick a safe first job
A good first agent job is repeated, read-heavy, reversible and easy to check. Examples: a morning digest of news on a topic you follow, a weekly summary of your own notes, a list of overdue items from a tracking sheet, a draft agenda from your calendar. Nothing leaves your hands without you.
Avoid first jobs that move money, change accounts, contact customers or touch private records about other people. Those can come later, in small steps, with approvals and your employer’s permission where work data is involved.
Least permission
Give the agent the narrowest access that works. Read-only instead of edit. One folder instead of the whole drive. One mailbox label instead of the whole inbox. Use a separate account or test space when you can. Remove access you no longer need.
Never paste passwords or secret keys into a chat or an instruction. Use the product’s own connection settings. At work, follow your employer’s policy and ask before connecting work accounts. Where patient, client, student or citizen information is involved, use only tools your employer has approved for it and follow your profession’s privacy rules.
Cost awareness
Running agents can cost more than chat, because they take many steps and read a lot of text. Costs usually grow with the number of steps, the amount read, repeated attempts and how often the job runs. Pricing differs by product and plan, so check the current help pages and billing settings.
Control it by limiting scope. Say how many items to handle, how many steps to take, and when to stop. Start with a small run, look at the usage, and then decide whether to schedule it. If your plan allows spending limits or alerts, turn them on.
Your first-run routine
Write the goal, tools, limits and approval points on one page. Run it once by hand on a small sample. Read the agent’s report and check results against the real sources. Change the instructions. Run it again. Only then consider scheduling it, and keep checking it on a regular day.
Worked example: a small charity manager in Nairobi has an agent read her own event calendar sheet each Monday and draft a list of reminders for her review. The sheet holds no personal details. The agent has read-only access to that one sheet, drafts only, and stops after fifty rows. After a month of correct drafts, she adds a second job. Trust grew from evidence.
Practice now
- Write a one-page plan for a first agent job: goal, tools, permissions, limits on steps and items, approval points and how you will check the result.
- List every account or file the job would touch and write the narrowest permission for each. Decide what to do if the job fails or runs too long.
Use any AI assistant you have. Do not paste private, client or patient information into it.
Check yourself
1. Which is a good first agent job?
2. What does least permission mean?
3. What should you do before scheduling an agent to run on its own?
Ask the AI tutor
Ask anything about this lesson, in any language. The tutor is an AI: it can be wrong, and it never needs your personal details.
AI-drafted. Checked by an automated validator and a separate AI reviewer; a person has not read every lesson. Translations and tutor answers are AI-made. Found a mistake? Tell us. AI can be wrong. General information only, not professional advice. Not affiliated with any AI company named. Current University is not accredited.