What tasks need to be given to AI urgently, and which ones should not be given in any case?

Now many people use artificial intelligence incorrectly. People give AI what they should do themselves, and they do what AI would do better. But here the question is how to understand what to give to AI and what not, how to prioritize, where to draw this line.

And to understand this, you need to understand one fundamental idea that came to us from economics. It can be shown in two graphs that explain how our returns or the effect of your efforts are arranged. And when you understand this, everything will fall into place.

Two laws from economics that explain everything

First Curve: Decreasing returns by Alfred Marshall

Each of your tasks falls into one of these two schedules. The first chart is the "diminishing returns" chart. The term was coined by Alfred Marshall. He is one of the founders of modern economic theory. And its essence is very simple. In most cases, the more resources you invest, the smaller the additional result.

That is, at some point, the effect of how much effort, time, and resources you put into a particular task begins to gradually decrease. For example, you are writing text. In the first hour of work, you have a draft ready. The structure is there, the thoughts are laid out, this is a huge progress. In the second hour, you are already starting to edit this text, improve the wording, and the text becomes better, but not as radically as after the first hour. In the third hour, you polish the details, change words, search for synonyms, the text has become even better, but the difference will already be almost invisible to the reader.

Accordingly, in the fourth, fifth, or sixth hour, you can rearrange commas and choose fonts, but the result will practically not change. The same goes for the presentation. You've made your slides, and the information is there. The most important thing is that there is logic. And then you start moving elements, changing shades of blue, and adding the perfect icon. It may take hours, but the audience won't notice.

If you look at the chart, then on the chart — this is a curve that first goes sharply up, and then comes to a plateau.

Second Curve: Increasing Returns by Brian Arthur

The second chart is exactly the "increasing returns" chart. It was prescribed by economist Brian Arthur. He studied technology companies, network effects, and innovation, and found that the opposite logic works in some situations. On the contrary, the more you invest, the more you get.

Example: negotiations with an important client. Here, in fact, every minute can radically change the result. One correct phrase and you have an investment of millions. Or a relationship. Each new effort to lift another person makes the next effort even more effective. One small action, a small improvement of 1%, gives much more effect in the future. On the chart — it is a curve that goes up and continues to go up.

How do I determine which zone your task is in?

So, everything that falls into the first chart, feel free to give AI. Everything that falls into the second one is your growth zone. And here you need to remember this. Every improvement you make may increase your chance of success.

Ask yourself two questions. First question. If I make this task 10% better, will something fundamentally change? If the answer is no, this is the first curve. If yes, it can change everything, the second one.

The next question is whether this task has an objective quality ceiling. The presentation has it. It's either clear or not. Negotiations don't. You can always negotiate on the best terms.

Another feature of the first curve that we give AI is repeatability. If you do something like this every week, every day is probably a routine. It can be delegated to the AI. The second curve indicates that the context is unique. Because every situation is a little different. You need to think again, apply judgment. There are no templates as such, because you need to turn on your head every time.

4R Framework: What are the four types of tasks to give AI right now?

Let's move on to practice. I created a simple system of four types of tasks. I call it 4R. These are four types of tasks that belong to the first curve, which can and should be delegated to AI.

1R-Overclocking: how to overcome the fear of a white sheet?

This is all where there is a fear of a white sheet, where you need to start, but it is not clear how. Lack of motivation, procrastination, procrastination that there is no understanding of what exactly needs to be done. So, the first draft of the text, the presentation plan, the structure of the material, the list of ideas for the project, the outline of the letter, the product concept — all these are classic tasks of the first curve. AI is not a substitute here, but it gives you the starting speed and acceleration.

It's like an airplane. Before you take off, you need to gain a certain speed so that the aerodynamics work, and the wings lift the car. You can't take off without acceleration. We all really need literally one idea, literally one sentence, why brainstorming is so effective, because it just starts this overclocking process.

In my experience, it's worth saying that a draft can be bad, and it almost always is. This is normal, it's even good, because his task is not to solve the problem for you, his task is just to start your thinking. And you know, I often have the opposite, when I come across the wrong ideas, I come up with the right idea.

I always try different models, I highly recommend it. Claude works better with text, it is deeper, more thoughtful. ChatGPT works better iteratively, i.e. when you test a lot of ideas quickly. Each model has its own view and angle.

2R-Research: How does AI speed up data collection?

The second R is reserch, information collection, research, and data retrieval. An AI agent is already a completely different level, because it has more time, it has much more ability to search for information, it can go to the Internet itself, it can open sites, read pages, click on links, and collect information from various sources. He acts here more like your assistant-resercher.

For example, I needed to understand what pains people who want to learn or implement automation have, and what stops them, what objections they have. I gave the agent promt, and he went to Reddit, according to thematic forums, tweets and comments, and collected such a huge report with specific quotes, links and a structure by topic. It would have taken me a couple of hours, here it happened in 10 minutes.

Another example is the research summary. When you need to quickly understand what is generally known about a topic, what data is available, what facts are available, and what research is available. This is especially useful when you're preparing for an important decision, or perhaps want to test your idea, or just want to know if it's worth digging deeper at all.

And once again, I want to emphasize that AI is a search accelerator, but not a source of truth. We trust AI only to collect information, it shows you where to look, it saves you hours of routine, but it's up to you to check and decide. Never accept what the AI says as fact without checking it.

3R-Ranking: how to turn chaos into structure?

Ranking is when you take a bunch of disparate information and sort it out. What is important, what is not, where they are related, where there are some patterns, where there are contradictions. Simply put, it is the transformation of chaos and all information into a structure.

Structuring notes, highlighting patterns in data, summarizing long documents, comparing options, prioritizing tasks, and grouping ideas by topic are all classic tasks of the first curve. And AI here can really save a huge amount of time.

Another powerful technique is cross-analysis. This is when you upload several different sources and ask them to find links between them. For example, three documents — find out what they have in common, where they contradict each other.

And the most unusual thing is multimodal analysis. AI accepts different formats, not just text. You can upload files, a link, a PDF document, an audio transcription, a table, all together, in different formats to the agent. And it can combine all this data into a single image. Such different formats, disparate information, processed into one single image, which can already be analyzed and made decisions. In other words, AI helps you see the structure in chaos.

4R Routine: what should I automate first?

The fourth R is routine. These are small repetitive tasks that you do the same every day. Something that does not require making decisions, but only execution. For example, formatting text, translating documents, transferring meanings between formats, clearing data, creating previews and cards, and composing identical emails. Automate everything that can be automated.

But here is a very important principle-don't get involved in automation if you don't understand the processes yourself. It is very important. If you want to learn automation, you first need to understand what you want to automate. If you don't understand this area, you'll be automating nonsense.

Three levels: from manual work to full automation

Such a chain works effectively here. First, you can do something with your hands, without AI. Then you learn to do it together with AI, simplify something. And you only then automate it. Automation is the transfer of your process to an algorithm or agent. If you don't understand the processes yourself, you simply won't be able to pass it correctly.

For example, how it works in practice. Let's say you run a blog. There is Telegram, there is Instagram, there are some other platforms. You write a post in Telegram, your thoughts, ideas, everything in detail. First level — you write manually. Take a post, shorten it, highlight key thoughts, write text for cards, create cards, upload them. This takes time.

The second level is connecting AI. Take the text of the post, ask the AI to split it into cards, and look for the wording of the product that best copes with this. You experiment, try different recipes, and change styles until you get the perfect process.

And then only the third level comes, when you already have everything, ready-made scripts, what works, what doesn't, how to write correctly, only then you already train the agent and pass this task. Or even make an application that will automate this task. Here is a ready-made automation of a specific task, you can use it and pass it on to others.

What AI will never be able to do for you?

Unlimited-impact tasks

When you have already mastered the 4R framework, you will free up time and resources for the second curve, for what cannot be delegated. And remember that everything that concerns the second curve is tasks with unlimited returns. Here you should understand that no matter how lazy it is, every real improvement, every more minute spent gives a disproportionately large result.

What can it be? Strategy — choosing directions, setting priorities, deciding what to do and what not to do. Making decisions — especially in conditions of uncertainty, when there is simply no right answer, when you need to take responsibility for yourself, and this is only a human task.

Working with people — understanding people's motivations, reading emotions, and building relationships. Negotiations — finding solutions that suit everyone, managing conflicts, reaching agreements. Creative — ideas that cannot be deduced logically, which appear on experience, unexpected connections, intuitive moments. This is only our human ability, and it needs to be developed.

AI can help you prepare for negotiations, gather information, and search for data about a person. But the negotiations themselves are you. AI can formulate 100 ideas, but it's up to you to choose the only one that works. AI can analyze the data, but it's up to you to decide what to do with it. AI can do the work for you, but it can't think for you.

How do I start using the 4R framework today?

I have prepared a special prompt assistant that will help you go all the way, from analyzing your tasks to specific instructions on what to delegate and how to do it. Just insert your weekly to-do list, and the AI will help you prioritize things according to the laws of decreasing and increasing returns. And then try to apply the 4R framework to your tasks.