How to Break Down Your Work into Tasks and Understand Which of Them Can Already Be Handed to Neural Networks

Breaking down your own tasks is the first step recommended by experts from the VFokus Mail material dated October 30, 2026. Below is a concrete procedure. It requires no special knowledge beyond the ability to work with spreadsheets and run Python scripts.

What you will need

• A list of all your work tasks over the past two weeks. A calendar, task tracker, or draft in notes will do.
• A spreadsheet (Excel, Google Sheets, or CSV).
• Access to any neural network via API (e.g., OpenAI, YandexGPT, GigaChat). The key is issued in the service's personal account. In the code below it is designated as YOUR_KEY.
• Python 3.9+ and the requests library. Installation: pip install requests.

Step 1. Export tasks to a spreadsheet

Open your calendar for two weeks. Write down every action that took more than 10 minutes. Format: name, average time in minutes, recurrence (daily, weekly, one-time). Example rows: "responding to standard client emails, 40 min, daily", "preparing a sales report, 60 min, weekly".

Do not filter. The more rows, the more accurate the picture. If a task repeats but you dislike it, still record it.

Step 2. Tag tasks by three attributes

Add three columns: "text/data work", "requires human communication", "requires non-standard solution". Put 1 or 0.

Tasks with a 1 in the first column and zeros in the other two are candidates for automation. These are primary document processing, standard correspondence, translations, scheduling — what the source calls routine.

Tasks with a 1 in the "communication" or "non-standard solution" column are better left to humans for now. That is where algorithms perform worse: task formulation, result verification, relationship building.

Step 3. Run candidates through the neural network

Take 5–10 candidate tasks. For each, formulate a prompt and send it to the API. Below is a working script. It reads a CSV with tasks and for each writes a short model response: whether it can be fully automated, partially, or not at all.

import csv
import requests

API_URL = "https://api.openai.com/v1/chat/completions"  # replace with your provider's URL
API_KEY = "YOUR_KEY"  # obtain from the service's personal account

with open("tasks.csv", newline="", encoding="utf-8") as f:
    reader = csv.DictReader(f)
    for row in reader:
        task = row["task"]
        prompt = (
            f"Task: {task}. Assess whether it can be fully or partially "
            "automated using a neural network. Answer in one sentence."
        )
        response = requests.post(
            API_URL,
            headers={"Authorization": f"Bearer {API_KEY}"},
            json={
                "model": "gpt-4o-mini",  # check the model name with your provider
                "messages": [{"role": "user", "content": prompt}],
                "temperature": 0,
            },
            timeout=30,
        )
        response.raise_for_status()
        answer = response.json()["choices"][0]["message"]["content"]
        print(f"{task} -> {answer}")

The file tasks.csv must have a task column. Example content:

task
Responding to a standard client email
Preparing a weekly report
Translating a short text from English

Run: python check_tasks.py. The script will output an assessment for each row. This is not the ultimate truth, but a quick way to see where the model is confident and where it is not.

Step 4. Test on a real task

Choose one task that the model called automatable. Do it manually and via the neural network. Compare time and quality. If the model's result requires more edits than manual work, postpone automation.

Repeat for three tasks. That is enough to understand where you personally gain.

Step 5. Create a learning plan

For the tasks that remain with you, write down the skills that strengthen them. The source indicates: analytics, management, communication, quality control. Add the ability to formulate prompts and critically evaluate responses.

Choose one skill and one course or project where you will apply it within a month. Do not try to compete with the algorithm in routine speed — master what complements it.

Typical mistakes and how to recognize them

Mistake 1. Tasks too large. "Managing a project" cannot be automated. Break it down into "collect status from three meetings". Sign: the model responds with generalities.

Mistake 2. Ignoring verification. You sent a task to the neural network and published the result without edits. Sign: the text contains factual errors or invented figures. Always check the output.

Mistake 3. Automating communication. Correspondence with a client via a template without a human can worsen relationships. Sign: the interlocutor replies more briefly or asks for a "real person". Leave such tasks to yourself.

Mistake 4. Key in code. You committed YOUR_KEY to a public repository. Sign: the service sends an email about suspicious activity. Store the key in an environment variable.

Mistake 5. Assessment without measurement. You decided the task is automatable but did not measure time before and after. Sign: a week later you are still doing it manually. Keep a simple log: task, manual time, AI time, number of edits.