What Is an AI Agent? A Plain-English Explanation With 12 Real Examples (2026)
What is an AI agent, explained for complete beginners in 2026: how it differs from a chatbot, what it does on its own, 12 real examples from everyday life, what it cannot do, the ten words you will hear, what it costs, how to stay safe, and how to try one free in ten minutes without a card.

You have used ChatGPT once or twice, maybe to write a birthday message. Now the news, your nephew and every app on your phone are talking about "AI agents", and nobody has explained what the word means, what the thing does, or whether it is for people like you.
This guide answers that in plain English. An AI agent is software you give a job to rather than a question, and the difference matters more than the jargon suggests. Below: the one-sentence definition, the test that tells a chatbot from an agent, what an agent actually does when you press go, 12 real examples, what it cannot do, the ten words you will hear, what it costs and how to try one free in ten minutes. Facts were checked on 8 October 2026 and every number has a named source.
Want to see one work before you read on? Try an agent on Jobbit: 100 free credits a day, no card, nothing to install.
How we wrote this guide
- Definitions from the people who build agents. Anthropic's published guide to agents is the reference for what the word means.
- Numbers from Ofcom and Gartner. UK usage from the regulator's Online Nation report, business adoption from the analyst firm.
- Examples we have run. Every example in the table is a task an agent can do today, not a demo.
- No jargon without a translation. Every technical word gets a plain sentence the first time it appears.
The one-sentence answer
An AI agent is a computer program that you give a goal, which then works out the steps itself, uses tools such as a web browser, files and email to carry them out, checks its own progress, and comes back to you when it is done or stuck.
Anthropic, the company behind Claude, puts it this way in its guide to building agents: agents are systems where the AI model "dynamically direct their own processes and tool usage". The important words are "direct their own". A chatbot waits for your next message. An agent decides what to do next.
Two numbers show why everyone is suddenly talking about this. Ofcom's Online Nation report, published in December 2025, counted 1.8 billion UK visits to ChatGPT in the first eight months of 2025, up from 368 million in the same months of 2024. And Gartner predicted in August 2025 that 40% of enterprise applications would include task-specific AI agents by the end of 2026, up from under 5% in 2025. The chatbots arrived first; the agents are the next step.
The taxi test: chatbot, assistant or agent?
Ask three pieces of software the same thing: "I need to get to Manchester Airport for a 6am flight on Friday."
- A chatbot explains your options: train times, the cost of a taxi, where to park. It answers and stops.
- An assistant does one thing when told: "book a taxi for 4am" and it books the taxi.
- An agent works the problem: checks the flight time, works out when you need to leave, compares a taxi with the train, books the one you prefer, puts it in your calendar and tells you what it did, asking you before it spends money.
Most products you have heard of now contain all three layers. ChatGPT, Claude, Gemini, Microsoft Copilot, Manus and Meta's Muse all started as chat and have added agent features, usually on their paid plans. Our guide on the difference between a chatbot and an AI agent goes through twelve side-by-side examples.
What an agent actually does when you press go
The loop is the same in every agent, and once you have seen it the technology stops being mysterious.
- It reads your goal and works out what "done" would look like.
- It makes a plan, usually a short list of steps.
- It uses tools. A web browser to look things up, a spreadsheet to add things up, your files to read from, email to send with, a code runner to build things.
- It checks the result of each step against the plan. If a website would not load or a number looks wrong, it tries another way.
- It reports back, and stops for your approval before anything that costs money or cannot be undone.
Our guide on how AI agents work opens up each step, including why the agent sometimes gets things wrong and why it is billed in credits.
12 real examples of what an AI agent can do
| # | You say | The agent does | Your time | Watch out for |
|---|---|---|---|---|
| 1 | "Find me a cheaper home insurance renewal than this letter" | Reads the letter, searches comparison sites, lists three quotes with the differences | 2 minutes to check | Confirm the cover is the same before switching |
| 2 | "Book a table for four on Saturday near the station" | Checks reviews, finds availability, holds a booking for your yes | 1 minute | It should ask before booking, not after |
| 3 | "Turn these 40 receipt photos into a spreadsheet" | Reads each photo, extracts date, shop and amount, totals by month | 5 minutes to spot-check | Blurred receipts need a second look |
| 4 | "Which of these five vacuum cleaners should I buy?" | Reads reviews and specs, builds a comparison, recommends one for your floors | 3 minutes | Ask it where each fact came from |
| 5 | "Write and send a complaint about this parking fine" | Drafts the letter with the right reference numbers and sends it when you approve | 5 minutes | Read it before it goes |
| 6 | "Plan a three-day trip to Lisbon in March under £600" | Builds an itinerary, finds flights and a hotel, holds options for approval | 10 minutes | Prices change; approve the same day |
| 7 | "Make a one-page website for the allotment society" | Writes the page, adds a contact form, puts it online on a web address | 15 minutes to review | Check the phone number and the opening times |
| 8 | "Chase the three invoices that are overdue" | Finds them, writes polite reminders in your tone, sends and logs them | 2 minutes | Set the tone once and it keeps it |
| 9 | "Read this 40-page contract and tell me what I am agreeing to" | Summarises it, lists the clauses that favour the other side, answers questions | 10 minutes | A summary is not legal advice |
| 10 | "Turn this recording of the meeting into minutes and actions" | Transcribes it, writes minutes, lists who does what by when | 5 minutes | Names it has not heard before may be misspelt |
| 11 | "Every Monday, send me a one-page report of last week's sales" | Builds the report once and runs it on a schedule without you | 10 minutes once | Check the first two runs |
| 12 | "Design a poster for the school fair" | Drafts three designs, makes your edits, exports a print-ready file | 10 minutes | Check dates and spelling on the final |
Notice the pattern in the third column. Your job is not to do the task; it is to check the result and give the yes. That is the whole change an agent makes, and it is why the "watch out" column exists. Our guide to 12 everyday jobs to hand an AI agent turns several of these into a first week.
What an AI agent cannot do
- Know things it has not been shown. An agent only knows what is on the web, in the files you give it and in what you type. It cannot read your mind or your filing cabinet.
- Be right every time. AI models sometimes state things that are not true, with complete confidence. OpenAI's own research paper on the subject, published in September 2025, explains that the way models are trained and tested rewards guessing over saying "I do not know". The agent's checking loop catches many of these, not all. You check the output; it does not check you.
- Act where it is not allowed. It cannot log in to your bank unless you let it, and it should ask before spending money. A growing share of websites also block agents outright, as our report on the web locking AI agents out explains.
- Replace judgement. It can draft the complaint; it cannot decide whether to fall out with your neighbour. Legal, medical and money decisions stay yours.
- Turn up with a van. When the job is physical, or needs a qualified person, an agent can only find and brief one. Jobbit was built around that gap: its agent does the work it can, and sends the rest as briefed, scoped jobs to vetted professionals in its human network, with payment held in escrow.
The ten words you will hear
- Model (or LLM, large language model): the AI engine that reads and writes language. Claude, GPT and Gemini are models.
- Prompt: what you type. A goal with the details is a good prompt; a one-line wish is not. Our guide on how to write prompts for AI agents has 30 examples.
- Token: the unit a model reads and writes in, roughly three quarters of a word. Prices are often set per token.
- Context window: how much the model can hold in mind at once. A long document can fill it.
- Tool: anything the agent can use beyond talking: a browser, a spreadsheet, email, a code runner.
- Browser agent (or computer use): an agent that works a website the way you would, clicking and typing. Jobbit lets you watch it and take over.
- Credits: the currency most agent services charge in. A simple task costs a few, a long research job many.
- Hallucination: a confident wrong answer.
- Agentic: the adjective. "Agentic AI" means AI that acts, not just answers. Our guide to what agentic AI is covers the five levels of autonomy.
- Human in the loop: you, checking and approving. The best agents are designed around it rather than around removing you.
The full AI agent glossary explains 60 terms if you want the rest.
What an AI agent costs
Less than people expect, and often nothing to start. ChatGPT made text chat unlimited on its free plan in August 2026, with its agent mode on paid plans, as Engadget reported. Claude's free plan includes web search, files and code, with its Chrome browser agent on the paid plans. Gemini, Copilot and Manus each have a free tier with daily limits, and Meta's Muse launched with a free tier in the United States in September 2026.
On Jobbit the Free plan gives 100 credits a day with no card, and Go is £7 a month for 2,000 credits and one small app kept live on your own domain. Our guide to how much AI agents cost compares the paid plans, and our round-up of free AI agents compares what each free plan actually allows.
Is it safe to use?
Safe enough for everyday tasks if you follow three rules. Never paste passwords, card numbers or other people's personal details into any AI tool. Keep the agent on "ask before you act" for anything involving money or sending messages. And check the output before you rely on it. Our guide on whether AI is safe to use has the full ten-point checklist and your rights under UK law.
How to try an AI agent in ten minutes
ChatGPT, Claude, Gemini, Microsoft Copilot, Manus and Meta's Muse all let you start with a chat. Jobbit starts with a job.
- Minute 1: open jobbit.uk and create a free account. No card.
- Minute 2: pick a task from the table above that you actually need done this week. Not a test. A real one.
- Minute 3: type it as you would say it to a capable new assistant: what you want, what good looks like, and anything it must not do.
- Minutes 4 to 8: watch it work. Jobbit shows you the plan and the browser as it goes, and you can stop it or take over at any point.
- Minute 9: check the result against your own knowledge. Ask it where a fact came from if you are unsure.
- Minute 10: give the yes, or say what to change. That is the whole skill.
If the job turns out to need a person, a plumber, a photographer, an accountant, the same agent briefs a vetted professional from the human network and holds your payment in escrow until the work is done.
Frequently asked questions
What is an AI agent in simple terms?
A program you give a goal to, which plans the steps, uses tools such as a web browser and your files to do them, checks its work and reports back. A chatbot answers a question; an agent finishes a job.
What is the difference between an AI agent and ChatGPT?
ChatGPT began as a chatbot: you type, it replies. It now has an agent mode on its paid plans that can browse and complete tasks. "Agent" describes what the software does, not which company made it.
Is an AI agent the same as agentic AI?
Nearly. Agentic AI is the adjective for AI that acts rather than only answers, and an AI agent is a product built that way. People use the two interchangeably.
Do I need to know how to code to use an AI agent?
No. You describe the job in ordinary language. The agent writes any code it needs itself, and the best ones show you what they are doing in plain words.
Can an AI agent spend my money?
Only if you let it. A well-designed agent asks before it pays for anything, sends a message or deletes something, and Jobbit's agent stops for your approval at those points.
Are AI agents free?
Most have a free tier with daily limits, and the fuller agent features are usually on paid plans of about £7 to £20 a month. Jobbit's Free plan gives 100 credits a day with no card.
How do I try an AI agent for free?
Create a free Jobbit account, give the agent one real task from your week, watch it work and check the result. 100 free credits a day, no card, and Go is £7 a month if you want an app of your own kept live.