AI chatbot in Telegram: 10 real tasks we gave an agent, with its replies

The short answer

An AI chatbot in Telegram is useful when it finishes the job inside the chat. You write one message in plain words. The agent picks the tool it needs: web search, a file, an image, a timer. Then it sends back the result.

We tested that on 28 October 2026. Ten requests went to a WAM agent on its default model, one per message, no setup between them. Every reply came back in under two minutes. The slowest was the printable file, at 106 seconds; most took 30 to 50. If you first want the basics of how an agent lives in Telegram, start with “An AI agent in Telegram that does the work”. Here we go straight to the work.

1. Rewrite a message so it lands well

The most common job is also the smallest. You have a blunt draft, and you need it polite without losing the point. You paste it and say what tone you want.

Two versions arrived in 31 seconds. The agent added the reason the other side cares about: a finished report instead of a draft.

2. Translate with the right level of formality

Translation is where a chat assistant earns its keep in a messenger. You copy a line from one chat and paste the result into another.

We asked for two languages and got four versions. The agent split each language into formal and informal address on its own, and explained one word choice.

3. Pull tasks and owners out of a team thread

Work chats bury decisions in banter. Paste or forward the thread and ask who owes what. Ours had seven messages from three people.

The useful part is the last block. The agent turned relative days into dates and spotted the one task with no owner. It also saw that Anna's test depends on Mark's numbers. In a group the same request works with a mention: the agent hears only messages addressed to it when /mention on is set.

4. Search the web and give the source

A model on its own knows what it saw in training. For anything current the agent runs a web search, which costs 1.5 magnets. We asked about something that changes often and is easy to check.

We checked the answer against python.org the same afternoon. The version and both dates matched, and all four PEPs appear in the official “What’s new” page. The reply took 39 seconds.

5. Draw a picture for a post

Pictures come from the same chat. On the default model one image costs 20 magnets. We asked for a simple post illustration.

It arrived 52 seconds after the request. The agent also left room at the top for a headline, which we never asked for. More prompts and the pictures they made are in “An AI image generator inside your chat: six prompts and the pictures they made”.

6. Make a file you can print

Some answers belong in a file. Ask for a file and the agent builds one and sends it to the chat as a document.

This was the longest task, 106 seconds. The Word file opens with a header line, “3-day work trip · 2 nights · one evening dinner”. Then come three sections with tick boxes, and the dinner outfit, shoes and accessories form one group. The same works the other way round: forward a PDF and ask questions about it, as in “Twelve things to ask an AI about a PDF”.

7. Set a reminder in plain words

Reminders are the clearest difference from a chat window you have to open. The agent writes first, at the time you set.

It converted “in 2 minutes” into a clock time in our time zone and arrived on schedule. One-off reminders like this work on Free. On Pro the agent also repeats a reminder on schedule, such as every Monday at 9.

8. Draft an email

The agent writes mail the way you would after a coffee. We asked only for the text and sent nothing.

Connect Gmail with one link in the chat and the same request ends with “send it”. The agent then sends from your mailbox and can sort the inbox first. How that access works without your password is covered in “Is an AI agent safe? 8 things to check before you let one into your chats”.

9. Split a shared bill

Three people and three receipts is a job for the agent: it does the sums in one message. Ask for the fewest transfers.

The sums are right: 240 + 96 + 60 = 396, a third is 132. Reply time: 28 seconds.

10. Read a link and turn it into a plan

The last request combined two jobs: read a page and plan an evening with it. We sent the Wikipedia article on the Pomodoro Technique.

Note the line about the time. Our message arrived after 19:00, so the agent moved the block to the next day and named the date. Then it offered the reminder that closes the loop.

How this differs from a regular Telegram bot

A typical Telegram bot does one job through a menu. A currency bot converts currencies. A support bot routes you through buttons to an operator. They are fast and predictable at that one job.

An AI agent works from the task itself. The ten requests above went to one contact, in one chat, with no commands. The agent decided where it needed search, a file, a picture or a timer. It also remembers you between chats: your projects, your tone, the people you work with. Memory is edited in plain words, “remember” and “forget”. The terms behind all this, from context window to MCP, are in “AI agent in Telegram: 18 terms in plain English”.

Each WAM agent is a separate Telegram bot with the name and avatar you choose. It starts in Telegram, and other messengers can be connected.

What to know before you start

For an important fact, open the sources the agent attaches. Our search reply came with links, so a look at the source took two minutes.

Try the same ten requests

Open @wamagentsbot, hit Start and name your agent. Creating it takes about a minute. Then paste any request from this page as it is written. The first one usually tells you which of the ten you will use daily.