The AI world spent this week announcing agents that can live on their own computers and keep working after you leave. OpenAI debuted its Dots at DevDay on Tuesday, while Meta launched Muse earlier this month.
The biggest difference between an AI chatbot and an AI agent is how it handles various tasks when left alone. So while a chatbot answers a question, an agent actually does a job such as checking something, keeping track of information, working across tools or coming back when something changes.
And some of the best jobs to give an agent are the ones that humans find incredibly boring.
Here are seven I would happily hand over — and a few I wouldn't.
1. Triage your inbox
The job: Instead of opening your inbox and working through everything yourself, have an AI sort out what actually deserves your attention.
It can summarize new messages, separate FYIs from emails that require action and prepare drafts for routine responses. You still approve anything before it goes out.
Try this prompt: Go through my unread emails from the last 24 hours. Give me a one-line summary of each, flag anything that needs my attention today and draft responses to the routine messages. Don't send anything without my approval.
This is also a good example of where connected AI becomes more useful than simply pasting an email into ChatGPT. So you'll want to connect your agent such as Gemini, ChatGPT or Muse to access your inbox, from there, the job becomes something it can actually help manage so you don't have to manually do it.
2. Watch a price so you don't have to
The job: This is one of my favorites because I'm always looking to save money for my family of five, but I don't have time to sit and watch prices drop. Give an agent a flight, hotel or product you're considering, tell it the price you're willing to pay and have it keep checking. Instead of opening the same page every morning, you hear about it when something actually changes.
Try this prompt: Check the price of [product/flight/hotel] regularly. Let me know if it drops below $[X] or if you find a comparable option for less.
This is where persistent agents start to feel wildly different from chatbots. The useful part is giving the AI responsibility for checking again later.
3. Prepare you for tomorrow
The job: Give AI the tedious Sunday-night task of figuring out what tomorrow looks like. An agent with access to your calendar and other connected information can potentially identify your meetings, pull together relevant background and give you a short briefing on what deserves your attention.
Try this prompt: Look at my calendar for tomorrow. For each meeting, tell me what it's about, what I should know beforehand and anything I need to prepare. Finish with the three things I need to prioritize tomorrow.
You can do a simpler version by pasting your calendar into a chatbot, but an agent becomes much more useful when it can gather the information itself.
4. Keep track of the things you always forget
The job: Subscription renewals. School deadlines. Appointments. Packages. Forms that need to be returned. That thing you said you'd remember three weeks ago and absolutely did not remember. These are terrible jobs for humans because none of them are particularly difficult not to mention, there are just too many of them.
Try this prompt: Help me keep track of the deadlines, appointments and renewals I give you. Tell me what's coming up this week and remind me before anything needs action.
5. Do the research you've been avoiding
The job: Don't ask AI to simply recommend a hotel. Give it the tedious comparison work. Tell an agent what you're looking for, give it your constraints and have it research the options before coming back with a shortlist.
Try this prompt: I'm planning a four-day trip to [city] with a $2,000 budget. Compare flight and hotel options based on total price, location and reviews. Give me your three best combinations, explain the trade-offs and tell me which one you'd choose. Don't book anything.
I've tried this same approach for buying a laptop, finding summer camps, comparing internet plans and literally researching almost any purchase that would normally leave me with 40 browser tabs open.
6. Run the household admin nobody wants to run
The job: Every household has a constantly changing list of things that need to happen: groceries, appointments, things to buy, forms to complete, activities to schedule and random tasks someone needs to remember. Instead of rebuilding that list in your head every day, let AI maintain it.
Try this prompt: Be my household admin. Keep a running list of the tasks, appointments, purchases and errands I give you. Organize them by urgency and tell me what needs to happen today, this week and later.
This doesn't mean giving an agent permission to buy things or make important decisions by itself. The useful part is offloading the organizational work while keeping yourself in control of the actions.
7. Give yourself a briefing instead of another feed
The job: Pick something you genuinely want to follow and have AI turn it into a short recurring briefing. It could be news, your favorite sports team, a company you're following for a job interview, an upcoming vacation destination or a subject you're trying to learn.
Try this prompt: Every morning at 8, give me a five-minute briefing on [topic]. Tell me what's new, why it matters and what changed since yesterday. Skip anything you've already told me unless there's an update.
The important part is that you shouldn't have to remember to ask. A persistent agent can turn the prompt into an ongoing job rather than another conversation you have to start yourself.
The easiest way to tell whether you actually need an agent
If you can accomplish the job with one prompt and one answer, you probably don't need an AI agent. For example, asking ChatGPT to create a meal plan isn't particularly agentic. Neither is asking it to rewrite your résumé or explain the stock market.
Agents become interesting when the job continues after the conversation does.
"Find me a cheap flight" is a chatbot prompt but "Keep checking this flight and tell me when it falls below $300" is an agent job.
OpenAI's newly announced Dots push this idea much further. Each Dot gets its own cloud computer and browser and is designed to continue working after you leave. That opens the door to much more ambitious jobs, particularly ones that involve research, multiple steps or ongoing work.
But I wouldn't rush out and upgrade solely because you suddenly feel like you need an autonomous AI worker.
Most people should start by figuring out which repetitive jobs they actually want AI to handle. You can already experiment with many of the ideas above using the AI tools you have now, even if some of the checking and follow-up still requires you to come back manually.
Once you find yourself wishing the AI would continue the job without you, that's when an always-on agent starts making sense. And you can start small with free agents like Meta Muse. Other options that are less expensive than ChatGPT Dots are ChatGPT Work or Gemini Spark, both available for $20/month or less.
What I would never hand over to an AI agent
Giving AI more autonomy also means being much more careful about where that autonomy ends.
I wouldn't give an agent passwords or 2FA codes, unrestricted permission to move money or the final say over medical decisions. I also wouldn't let one independently sign contracts, submit legal documents or publish something publicly in my name.
And for anything involving purchases, messages or consequential account changes, I want a human approval step before the action happens. The goal isn't to give AI control of your life, just the boring parts.
How do you use AI agents? What's your favorite? Let me know in the comments.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom's Guide
