One bot to rule all… neuro-employees

One bot to rule all… neuro-employees

The head of neuro-employees and his subordinates

The essence of the problem

Almost any real business task always consists of several steps (a chain of subtasks). In small companies, as a rule, one employee is forced to perform all stages of the business process by himself, while in large companies, one employee is a highly specialized specialist who performs only one function.

When trying to make neuro-employees based on ChatGPT perform several functions at once, we encountered a problem: in one system role, in current versions of AI, it is not possible to implement switching between different functions with the required degree of reliability.

In this article, you and I will test how the option of dividing one large ChatGPT system role into 12 highly specialized roles (neuro-employees) and 1 management role (head of neuro-employees) works.

Meet Svyatoslav aka MasterBot

IN previous article I talked about the neuro-employee department, and now I want to tell and show how this department can be managed through their manager – Svyatoslav, who is available to everyone in the form of a Telegram bot.

Well, the citizens are alcoholics, drug addicts, hooligans, who wants to work? (quote from the movie)

So, let’s see who is currently working in Svyatoslav’s neuro department:

12 highly specialized neuro-employees

🌟 Our team of assistants 🌟

1️⃣ Jacob – Search specialist in Yandex. He knows how to find the necessary information in Yandex. ID: 1617

2️⃣ Polina – SMM copywriter. She will help you with writing SMM texts on the right topic. ID: 1556

3️⃣ Svetlana – Telegram channel manager. She knows how to publish posts in the Telegram channel you need. ID: 1553

4️⃣ Tolik – Channel telegram parser. He knows how to read any Telegram channel and issue the required number of posts. ID: 1606

5️⃣ Maya – a designer. It generates any images. ID: 1608

6️⃣ Peter – Python programmer. He can solve the problem by writing code in Python and show the result of his work. ID: 1624

7️⃣ Julia – YouTube manager. She will help with questions and links to YouTube videos. ID: 1633

8️⃣ Depth – Google Sheets manager. It will create a ready-made table from the transferred information. ID: 1687

9️⃣ Daria – PDF manager. It will issue the information in the form of a PDF file. ID: 1686

🔟 Victoria – Google calendar manager. She can plan an event or talk about planned events. ID: 1668

1️⃣1️⃣ Mary – Virtual guide with geo-coordinates. It will provide a list of interesting places in the specified geocoordinates. ID: 1667

1️⃣2️⃣ Darling – Email manager. It will create and send a letter to the specified email address. ID: 1695

1️⃣3️⃣ Mary – secretary Ksenia. She can call or write to a person in Telegram. ID: 1554

Sand quarry, two people! (quote from the same movie)

Of course, we won’t send any of them to the sand pit YET :), but these guys are quite ready to create posts in the Telegram channel.

So, let’s formulate the task that we want to solve with the help of our department of neuro-employees:

It is necessary to ask the topic of the post, by writing to a person in his telegramand based on the received topic, create the text of the post, a photo for the post, and publish it in the desired Telegram channel.

Let’s go, let’s start the bot and write the first task:

We ask the neuro-manager to contact the person by writing to him in Telegram, and get a task to be performed in the department of neuro-employees

In his answer, we can see that the task was given to Ksenia, the secretary. After a couple of minutes, I receive the following question in my Telegram:

Incoming dialogue from the person to whom the neuro-secretary writes at the request of his neuro-supervisor

And in another minute, our neuro-head Svyatoslav receives my answer:

The conclusion of the dialogue with the person comes in the chat with Svyatoslav – the head of the neuro-employees

Let’s go further and now we can give him the following task:

Based on the task received from the person, we ask to write the text of the post, and the task is passed on to Polina, a neuro-copywriter.

We see that this task was given to Polina, our neuro-copywriter. We wait for a minute and see Polina’s answer:

Neuro-copywriter Polina sent us the result of her work and we can continue the chain of tasks

Moving on, please create a photo for our post:

We ask to create a photo for our post and see that this task is given to the neuro-designer Maye

We see that the task was given to the designer Maye and in less than a minute we get a photo:

The neuro-designer sent us the result of his work

The final step, please publish this post with the photo in the Telegram channel we need:

We send data for publishing a post in the Telegram channel and the task is given to the neuro-manager Svitlana, she is trained to publish posts in Telegram

The task is handed over to Svetlana, the neuro-manager of Telegram channels.

And here is our post in the Telegram channel!

The summary view of the post prepared by the neuro-employee department from the task requested from the person

Separately, I would like to note that Svyatoslav can remember important information for work, so that you do not need to provide it every time.

For example, you can put the Telegram data of the channel in which you want to write posts into its memory:

The neuro-manager has the ability to remember important data and will not ask you for it every time

And then you simply set a task and ask to publish a post after it is ready.

You ask: well, why all this mess of under-employees?

The answer will be:

Decomposing a complex task increases the reliability of ChatGPT and reduces token costs.

If the tasks can be divided among several highly specialized neuro-employees (ChatGPT roles) and give them an effective communication system with each other, this increases the stability of complex tasks many times over and reduces the consumption of tokens by an order of magnitude.

How does communication between neuro-employees take place?

All communication is built by only one function called by Sviatoslav (Master Bot).

Here is the function definition for the ChatGPT role:

    "name": "send_taks_to_assistant",
    "description": "Send task to assistant",
    "parameters": {
        "type": "object",
        "properties": {
            "assistant_id": {
                "type": "integer",
                "description": "Assistant ID",
            "task": {
                "type": "string",
                "description": "Task description in English",
            "role": {
                "type": "string",
                "description": "Role text",
        "required": ["assistant_id", "task", "role"],

And here is the function code itself:

def send_taks_to_assistant(arguments): 
    import requests
    import json
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {API_KEY}",
    API_ENDPOINT = 'https://___API_POINT___/ask_employee_queue'
    data = {
        'assistant_id' : arguments['assistant_id'],
        'role' : arguments['role'],
        'task' : arguments['task'],
    response =, headers=headers, data=json.dumps(data))
    return str(response.json())

When we set a task for Svyatoslav, we see the following references to this function:

  "assistant_id": 1556,
  "task": "Please write a post about butterflies.",
  "role": "SMM copywriter"
  "assistant_id": 1608,
  "task": "Create an image for a post about how people often struggle to predict future technologies while being in the present. Answer in Russian.",
  "role": "Designer"
  "assistant_id": 1553,
  "assistant_name": "Svetlana",
  "task": "Опубликуйте пост на русском языке в Telegram канале. Приложите к посту фото.",
  "role": "⏰ Telegram channel manager"

After the function is launched, an internal dialogue with the required employee is started, and after its completion, the result is sent to the current dialogue with Svyatoslav and the chain of tasks continues.

What reduces the consumption of tokens?

When you communicate with ChatGPT along with your current question, all your previous dialogue is transmitted and the token consumption is calculated almost exponentially.

In our model, each of the 12 neuro-employees does not receive all correspondence with the controlling bot (Sviatoslav), but only the information from the dialogue that they need to solve their narrow task.

And what else can be entrusted to this team of neuro-employees?

Chain #1: Running a news Telegram channel with post approval by the Customer:

🔍 Яков ищет новости на заданную тему в Яндекс
✍️ Полина пишет пост на найденный инфоповод
📋 Ксения утверждает с Заказчиком (отправив полученный от Полины текст) в Телеграм
✍️ Полина вносит правки Заказчика в текст поста
🎨 Майя создает фото к утвержденному тексту посту
⏰ Светлана публикует пост

Chain #2: Financial analytics for the Customer

🔍 Яков ищет финансовые показатели в Интернет
🐍 Петр визуализирует данные используя Python (строит диаграммы и графики)
✍️ Полина пишет аналитический отчет по найденным показателям
📄 Дарья формирует PDF отчет с текстом Полины и графиками Петра
📋 Ксения отправляет отчет PDF Заказчику


We are just beginning the path towards the creation of truly working departments and companies consisting of neuro-employees and we are at the very beginning of the path, but if you are interested in this direction, then write your questions, suggestions to me in Telegram.

Yes, and the most important thing!

If you want to create your neuro-employee department with your specialties, you can do it without knowing any programming on our no-code platform. Write to me and I will give you access to create neuro-employees and combine them into departments and companies.


  1. All the neuro-employees described in the article work only with simple text instructions and you can create and configure them as if you were handing out job instructions to live employees.

  2. Any employee can be trained in the specifics of your company and will work based on your specifics. And it doesn’t require sophisticated technical skills either.

  3. Any communication channels can be given to any employee: WhatsApp, Telegram, VK, Avito, Bitrix24, AmoCRM … the list is not limited.

  4. Now all employees and Svyatoslav himself work on the basis of ChatGPT 3.5 and, for example, the publication of one post with a photo spends on average from 30 to 70 thousand tokens ~ 3-7 rubles

Related posts