AI Agent
Team

A small business needs several roles at once: someone who reads the numbers, the market and trends, someone who writes, someone who makes pictures, someone who publishes, someone who edits video. One all-purpose assistant mixes the roles. I built a team of eight AI agents on my own server, each with its own role, its own room in Telegram and a written list of what it never does

The marketer sends a daily brief: 80 briefs in 91 days. Every morning the analyst scans five industry chats and returns 20 live questions and 10 topics. The copywriter writes, the designer shows the idea first and draws after approval, the publisher pushes posts to six platforms, the video pipeline makes subtitles, and a night watcher checks the agents themselves. The same set of roles runs for a clinic owner, and personal assistants on the same core run for four other people

For a similar solution for your business, see Custom AI Agent Development

What You Get

If you need a marketer, a copywriter, a designer and an analyst but hiring four people is too much, the agent team takes those roles. You put a task in the right room by text or voice, and the agents never message each other or publish anything: the work moves through you, and the final word is always yours

How It Works

The Telegram group is split into topics, like folders. Each agent listens to one topic and only to the people you allowed. Anyone else gets silence

In the morning the analyst scans industry chats and returns live questions, topics and hot threads. The marketer adds yesterday’s posts, what worked, a trend from the web, a recommendation and three questions for you

The copywriter asks, offers three headlines, writes the post and edits it with you in the chat. The designer describes the picture in words, you approve, and only then does it draw. One post — one prompt — one file

An approved post goes to the platforms through the content factory, a phone video gets subtitles from one word. No agent publishes on its own — only on your button

Every night the watcher checks the agents and sends a card with buttons if something broke. Every agent has a written list of what it never does: the marketer writes no final texts, the copywriter draws nothing, the designer never draws without approval

The agents do not message each other: work is passed through you. That is a safety feature, not a limitation: nothing goes further without a person

Tools

• Claude

  the brain of every agent, different models for different roles

• Telegram

  a group with topics, one per agent

• Whisper

  understands voice tasks

• Image models

  the designer draws after approval

• Notion

  content plan

• Linux server

  a service per agent, the watcher on schedule

Result

Eight agents cover the roles of marketer, analyst, copywriter, designer, publisher and editor, and you stay the manager: you set tasks and approve results. Hours saved I have not measured and do not promise

The agents never message each other or publish on their own:
the work moves through you

I want the same