What AI agents actually are (beyond the hype)

AI agents – buzzword of the year?

There’s a question I’ve been thinking about for a long time: What actually are AI agents?

They’re everywhere. Technology and consulting firms keep feeding us predictions about how agents will transform the world of work. Headlines speak of agents already replacing entire professions.

And yet – I barely see real AI agents in the wild.

Especially not at those consulting firms, which in many cases are using nothing more than Copilot or ChatGPT. Sure, if you’re working with tools like Cursor or Claude Code, you’re already seeing genuine agentic workflows. The same goes for building automations in platforms like n8n. But an AI agent that works fully autonomously and can genuinely replace someone’s job? The reality looks very different.

I’ve wrestled with the word “agent” itself for a while. In our Superpowered Professional framework, we work with five levels of AI tool usage. And most people – when they talk about AI agents – are really describing assistants they’ve set up inside standard chat applications.

That’s why I kept thinking: the truly autonomous agents – the ones that make decisions and operate independently – I simply don’t see them with our clients.

How do you explain it to people?

In a recent conversation, we arrived at a simple but powerful insight:

“It doesn’t matter what an agent is anymore. People want to automate something and they don’t care what it’s called.”

The public has given up on precise definitions – either they get it, or they’ve reached the point where it feels too late to ask. And that’s exactly it.

My approach to agents: PACT

I now see everything – whether it’s an assistant or an agent – as a text file. One that describes what the agent should do, what context it should work with, and what tools it has at its disposal.

That’s why I built the PACT framework, which has completely changed the way I work:

🔥 Projects – folders containing information about specific activities (a newsletter, a home office renovation, whatever it is)

🔥 Agents – long prompts describing what the AI should do (an agent for building agents, an agent for processing transcripts…)

🔥 Context – my goals, principles, and knowledge base. This, alongside prompting, will be the most important thing of all.

🔥 Tools – the instruments agents use (pulling data from the web, voice-to-text transcription, video editing)

That’s it.

Once you combine these four elements, you have an AI capable of doing remarkable things for you. And it’s almost irrelevant which tool you use. Better tools handle more advanced tasks – that’s true. But even if you stick to basic chat-based AI and get your prompts and context properly dialed in, AI will serve you far better than it serves most people.

FD