What is an AI agent?
An AI agent is software that is given a goal and carries out the steps to reach it, using tools such as a browser, files and apps. It differs from a chatbot in one way that matters: it does the work instead of describing it. Ask it to find three suppliers and compare their prices, and it searches, reads, compares and hands you the spreadsheet.
The word agent means someone who acts on your behalf, and that is a useful test. Software that can only reply in a chat window is an assistant. Software that can open a website, fill in a form, build a file and come back when it has finished or needs a decision is an agent.
- An AI agent is given a goal and carries out the steps itself.
- It has a role, its own computer, access to the apps you choose and a memory.
- It starts from a message, a schedule or an event in a connected app.
- It asks before it sends, pays, posts or changes a record.
- It cannot take responsibility. A person reviews the work and decides.
What does an AI agent have?
A working agent has four things: a role, a computer, access to your apps and a memory. Take away any one and it falls back to being a chat window. On Monmo you set each of them yourself, much as you would set up a new member of staff on their first day.
| What it has | What that means on Monmo |
|---|---|
| A role | A job you describe in plain words, such as running the support inbox. The role says what to do and what to leave to you. |
| A computer | Its own desktop and browser on your team's cloud computer, with office, media and developer tools. It can use websites the way a person does. |
| Access | The apps you connect through your own accounts, from more than 1,700. You decide which apps each agent may use. |
| Memory | Your preferences and the background you give it, kept from one task to the next so you do not explain twice. |
How does an AI agent work?
It takes a task, works through it and brings back a result. On Monmo, work starts in one of three ways: you send a message or a voice note, a routine comes due, such as every weekday at 7:30, or something happens in a connected app, such as a new email. Work goes on after you close the app.
Along the way it reads, searches, compares, writes and builds files without asking. When the next step would send a message, spend money, post, publish or change something in your apps, it stops and shows a card, and you allow or decline. What comes back is a document, spreadsheet, deck, image, video or website, kept in the Library, or an update made in your apps.
How is an agent different from a chatbot and from an automation?
A chatbot waits for your question and answers it. An automation repeats one fixed rule. An agent works toward a goal and decides the steps as it goes. The three overlap at the edges, but the plain versions behave very differently when the work is a job and not a question.
- A chatbot gives you words. You read the answer, then do the work yourself.
- An automation gives you one step, every time, exactly as built. It cannot handle a case nobody planned for.
- An agent gives you the finished job. It reads the situation, chooses the steps, uses your apps and asks you before it acts on your behalf.
What can an AI agent do?
Most desk jobs that a capable person could do on a computer from clear instructions. The closest fit is work that comes round again and again, lives in apps and files, and produces something you can check. Results depend on the task and the sites involved, so review work that matters.
- Sort an inbox each weekday morning and draft the replies
- Research a market and write it up with sources
- Build a twelve-month budget with scenarios in a spreadsheet
- Turn a report into a ten-slide deck
- Reconcile bank lines and chase overdue invoices
- Plan a month of social posts with graphics
- Update the CRM after sales calls and report the pipeline
- Build a launch page and publish it once you approve
What can an AI agent not do?
It cannot take responsibility, and it cannot do anything away from a computer. Phone calls, meetings in person, physical work, signatures and legal or regulated sign-off stay with people. So do decisions that need accountability: pricing, hiring, firing, strategy and negotiation. An agent can prepare all of these, but a person decides.
An agent can also be wrong. It may misread a document, choose a weak source or misunderstand what you meant, and a website may not behave the same way twice. That is why it asks before it acts for you, and why its work deserves the same review you would give a new hire.
How do you start with an AI agent?
Begin with a single agent and a single job. Describe the role as you would to a person, connect only the apps that job needs, and set a first task whose result you can check. When it comes back right, agree a routine so the work happens without your asking. A second job is the moment for a second agent.