I have an AI agent working in my business. It is a character I call Jes, my business assistant. She helps me with everything related to my work, my website, creative projects and personal matters: she collects ideas, handles bookings and meetings in my calendar, and even receives inquiries and comments from my website. She is a full business assistant that I built first of all for myself, and she helps me every day

How an AI agent differs from a chatbot
A regular chatbot answers one question and forgets the conversation. An AI agent lives longer than one question: it remembers earlier conversations and my profile, runs jobs on a schedule on its own and hands tasks to other programs. I write to Jes in Telegram or send a voice note, and speech recognition turns it into a command
Jes has been working for me for 134 days. Over that time a clear set of tasks has formed that she takes on completely, and a set of decisions that stay with me
What an AI agent takes off your plate
First, scheduled work. Jes now runs 10 jobs that repeat and get done on their own: morning checks, reports, reminders. I don't need to keep them in my head or set the same task again every time
Second, teamwork. Jes hands tasks to 7 other bots, such as a copywriter and a designer, and comes back to me only when there is a result. Since July she has handed over 165 tasks this way
Third, small jobs that need attention rather than expertise: remind a client, check a status, follow up on a question three days later. Each one takes a minute, but together they eat the whole day
Fourth, the website. Jes receives inquiries and comments and prepares posts: the Russian text, an English version, a cover image and photos

What stays with a human
The main worry with an AI agent is that it will do something that can't be undone. That is why Jes has safety rules: deleting data, spending money, messaging clients and publishing always wait for my yes. I receive every finished post for approval and can send it back for a rewrite, ask for a different image or cancel it. Until I press Publish, the material stays a draft
An approval request comes to me in Telegram as a card: what the agent is about to do, where, and its status. I press Yes or Edit, and only then does the action go ahead
If the main AI model fails, the backup model continues on its own. And after my corrections the agent writes new rules for itself so it doesn't repeat the mistake
How your own agent differs from ChatGPT
ChatGPT is a shared service: it runs on someone else's servers, and your chats are stored there too. Your own agent works differently: it runs on your server or computer, the data stays with you, and the rules are written for your business. Its memory, the knowledge about you and your work that it builds up every day, belongs to you and can be moved to another system. Jes, for example, lives on my server as a 24/7 service that restarts itself after a crash
Where a business should start
Start not with choosing a model but with a list of jobs that repeat every day and eat people's time. Not “automate everything” but, for example, “check new inquiries every morning and send a short summary”. A good first candidate is a task where it is clear what comes in and what should come out. Then decide what the agent does on its own and what only after approval, and write it down like a job description for a new person on the team
Building an agent like this takes one to two weeks: setup, memory, approval rules, schedule and testing on your data. You can start with one job and add new ones as you go
You don't need to install a separate app for this: the agent works where you already work. You give it tasks in Telegram, it talks to clients in messengers, and it takes emails, meetings and files from your mail, calendar and cloud drive
I described how I build agents like this for business tasks on the page about AI agent development for business