# Pick your first AI agent job

> Researched. Small business · Beginner · Operations. Updated 2026-11-02.

Canonical page: https://www.aisetupsthatwork.com/setups/pick-your-first-ai-agent-job/

## What it does
Score your weekly tasks for time, repetition and risk, and pick one safe first job for an AI agent, with a 30-day test plan and a clear way to measure if it helped.

## Who it's for
Owners who keep hearing about AI agents but do not know which job to hand over first.

## Tools
- ChatGPT or Claude
- A list of your weekly tasks

## Steps
1. Write down every task you do more than once a week, with rough minutes each time.
2. Paste the list into the prompt below. It scores each task for time saved, repetition and risk.
3. Pick the top task that is low risk. Drafting, sorting and summarizing are safer first jobs than sending, paying or deleting.
4. Before connecting anything, remember OpenAI says a signed-in agent can access sensitive data and perform actions. Keep approval on for anything that leaves the business.
5. Run it for 30 days. Track minutes saved and every mistake.
6. Add a second job only after the first one runs a month without surprises.

## Copy-paste prompt
Adapt bracketed fields to your situation.
```text
Help me pick the first job to hand to an AI agent.

My weekly tasks (task, minutes each time, times per week): [PASTE LIST]

For each task, score 1 to 5:
- Time it eats per week
- How repetitive it is
- Risk if the AI gets it wrong (5 = very risky)

Then:
1. Rank the tasks by (time + repetition) minus risk.
2. Recommend one first job and say why.
3. Write a 30-day test plan: what the agent does, what I approve, and what I measure each week.

Use only my list. Do not suggest tools I did not mention.
```

## Guardrails
- What it touches: Chat input you provide (ChatGPT or Claude; A list of your weekly tasks).
- Needs your OK: Send, pay, publish, and account changes stay with you. Treat agent output as drafts first.
- Safety level: Read-only
- Starting with something risky, like payments, because it saves the most time.
- Connecting every account at once.
- Skipping the measuring, then not knowing if it helped.

## Tips
- The best first job is boring, frequent and easy to check.
- Measure minutes saved; it keeps the decision honest.
- One job running well beats five half-set-up agents.

## Estimates (labeled, not measured)
- Time saved: About 30 minutes to plan.
- Money saved: Not stated in sources.
- Tool cost: Free tiers can plan it. Agents that act may need a paid plan; prices: Not stated in sources.

## Sources
Summaries use the public sources below, not independent verification.
- OpenAI: https://help.openai.com/en/articles/11752874-chatgpt-agent (ChatGPT agent: when signed into sites or apps it can access sensitive data like emails, files or account settings and perform actions (read 2026-11-02 from a search result excerpt of the page))
- Anthropic: https://support.claude.com/en/articles/10166901-use-google-workspace-connectors (Use Google Workspace connectors: Gmail drafting, approval before send, reply or forward by default (read 2026-10-31 through a page fetch tool, not a direct browser open))

## Get the free starter pack
Five beginner copy-paste AI setups, free by email: https://www.aisetupsthatwork.com/starter-pack/?utm_source=ai-agent&utm_medium=markdown&utm_campaign=setup-pick-your-first-ai-agent-job

Related:
- [Set guardrails before an AI agent gets access](https://www.aisetupsthatwork.com/setups/set-guardrails-before-giving-an-ai-agent-access/index.md)
- [Make your AI agent stop for approval](https://www.aisetupsthatwork.com/setups/make-your-ai-agent-stop-for-approval-before-it-sends/index.md)
- [Morning inbox briefing and multi-inbox triage](https://www.aisetupsthatwork.com/setups/inbox-morning-triage/index.md)