The 4 Levels of Using AI: From Chatting to Agents

Preface

Hey, this is Leon. This is still the newsletter you subscribed to. I currently experiment with substack as a new service for my newsletter. This is why this article is sent from another E-Mail. That being said, let’s get into the article…

The AI Agent hype

LinkedIn is full of post like “AI will take your job” and “Use this Team of AI Agents to automate your life”.

So much, in fact, that I long was discouraged to engage with the topic at a more profound level at all.

— Now I do, so let me know if there is a specific topic you would like to read about —

The truth is, AI still has severe Limitations. I suggest we are somewhere around the peak of inflated expectations.

Different levels of working with AI come with different chances and limitations. Here comes an overview.

The 4 Levels of using AI:

This overview should help you to decide in as far different levels of AI solutions are suited for the problem you want to solve.

1.

Using the Chat Interface:
You just prompt into the Chat Interface. This is what most people do.

Just using the chat interface makes sense for tasks that are not recurring. To get optimal outputs, you need complex prompts. Building these prompts takes time.

The biggest potential lies in outsourcing recurring tasks. Here you don’t want to prompt again each time. This brings us to the next level.

2.

Building an “AI Assistant”:
You create a complex prompt once and save this prompt as a “project” (ChatGPT) or “new Agent“ (Microsoft Copilot). Every time you give an input, this prompt will be applied. You’ll receive the tailored output.

Some Ecosystems like Microsofts Copilot or Googles Gemini allow you to connect tools like their E-Mail or Calendar Software. Your Assistent then can write E-Mails or schedule events for you.

This is where the biggest return on invested time lies for personal productivity.

You don’t lose time prompting, but you get a profound output. One drawback is that you still need to act proactive and provide the input.

This is where automations can take over.

3.

Building an AI Automation:
You create a strict logic for AI to follow. When X happens, always respond with Y. AI has only the autonomy needed to solve the very specific steps of a well documented workflow.

This may come with the biggest ROI for businesses. It’s unpractical for personal use because building it is only efficient to solve a large-scale operation.

Processes must be recurring, follow a strict decision logic, must be perfectly mapped and extensively edge case tested or not too sensitive (i.e. if the Output is of poor quality nothing terrible happens).

At this level you also need workflow automation and orchestration tools like n8n, Lindy or Copilot Studio.

To have more flexible solutions to a bunch of problems without needing the human, we can take a look at AI Agents.

4.

Building an AI Agent:
AI automatically solves tasks autonomously with access to a great variety of tools and data. This is what everybody talks about.

In theory, this is the best way to use AI. You don’t need a human in the loop to prompt, but AI can deal with a great variety of problems at the same time, and those problem solutions must not be perfectly planned.

In reality, this does not work (yet). The more autonomy AI has to work on problems, the faster it gets lost, the more you need a human in the loop.

For now, the more complex the problems that you want to work on, the more refinement is needed by a human.

For personal productivity, try to work on what I called the “AI Assistant” level. Let me know If you would like hands on use case examples.

Let me also know how you liked the E-Mail format compared to the original one. You can vote here:

Until next Sunday!

Thanks for reading my newsletter! Subscribe to not miss out on new articles.