I previously wrote about some of my experiences with Agentic AI. Lately, a few new impressions have formed. They are no longer just observations about a tool. They are observations about what happens to my own mind while I work with it.

Three things stand out to me so far. I may add more later.

1. The intellectual demands on people are changing

In complex intellectual work, the most important ability used to be concentration. You stayed with one problem for a long time and kept digging.

It was never clear that multitasking offered an advantage. Working on several things at once usually meant being distracted again and again. Better to follow one direction deeply than to keep switching among several problems.

Agentic AI changes that way of working.

AI can advance many tasks at once and keep several work pipelines alive for you. If you drew the process, the old version might look like a few long, narrow tracks, each extending forward. Now many tracks can run in parallel. The shape of the work is changing from a handful of long lines into an expanding plane.

That changes the human role as well.

You still need to drill into a problem, but you must also decide which pipeline deserves to continue, which one has gone off course, how the pipelines depend on one another, when to pause or merge them, and when to open a new one.

AI may not lower the intellectual demand. It may change the structure of what intelligence asks of us.

That new structure requires people to switch quickly among lines of thought, recover the context after each switch, and return to deep concentration. You must let AI keep working on local problems while you step back to manage dependencies, assess results, notice deviations and choose where to go next.

Intellectual work used to resemble digging steadily through one tunnel. Now it is more like opening N tunnels at once. You take part in the digging, but you also move rapidly between them. Each time you enter a tunnel, you need to know at once how far it has gone, what it has hit and where it should go next.

That is a different ability from going deep inside a single pipeline.

2. Caught in AI’s "Star-Sucking Skill"

This second feeling is hard to describe precisely. For now, I call it being caught in AI’s "Star-Sucking Skill."

AI executes quickly. It can turn an idea into a plan, a piece of code, an essay or some other concrete result almost at once. The effect is especially obvious for people who have many ideas. Thoughts that once had to remain in the mind, or wait in a queue to be realised, can now be pulled out one by one and advanced through different pipelines at the same time.

It feels good, of course. It greatly expands what one person can do. Work that once moved slowly can now proceed on several fronts at once.

At the same time, I sometimes feel as though my brain has been hollowed out.

AI seems to keep drawing ideas out of my mind and turning them into reality at speed. On one side, I have become much more capable. On the other, I do not know whether externalising my thoughts at this intensity is draining the ability to think from me.

I cannot yet tell whether this is simply the fatigue of mobilising many ideas so quickly, or whether I have handed too much of the intermediate thinking to AI and stopped exercising some of my own abilities. Both may be true.

After I give an idea to AI, am I still thinking with it, or only waiting for the result? Once it finishes, have I actually absorbed the process? If the body of work keeps growing but never re-enters my own cognitive structure, then only the scale of the output has grown. My ability may not have grown with it.

That is perhaps the most contradictory part of using Agentic AI. It extends my ability while seeming to consume it from the other direction.

3. Breadth can be an advantage, or an illusion

The third feeling concerns the breadth of a person’s thinking.

Agentic AI may amplify someone who already has broad interests and can think across distant subjects. Such a person can pose questions in more directions, notice connections between fields and open more pipelines worth exploring.

My tentative view is that AI can compensate, to a degree, for limited depth in a local investigation. Breadth accumulated over years, along with the instinct to find worthwhile problems, is much harder to acquire. That may make breadth particularly valuable in the Agentic AI era.

But there is a trap.

AI can easily create an illusion of greater ability. In any direction, it can rapidly produce terminology, arguments, plans and results. A shallow idea can emerge looking thoroughly researched.

The depth may exist only in the AI’s output, not in the person using it. You can call it up without being able to judge it. You can display it without being able to restate, correct or advance it yourself.

By "breadth," I do not mean knowing many terms or briefly touching every subject. It must include at least a few learning paths you have genuinely travelled. Only by going deep yourself do you learn what depth looks like, where mistakes tend to occur and which results deserve trust. That experience lets you tell whether AI has supplied real reinforcement or merely the outward shape of depth.

When someone combines breadth with their own learning paths and standards of judgement, AI can amplify them enormously. When breadth means only skimming one subject after another, AI can package that shallowness more completely and make a person believe their ability has improved across the board.

Still observing

The more I use Agentic AI, the less I think it simply lowers the threshold for intellectual work. More precisely, it moves the threshold.

Sustained concentration still matters, but it is no longer enough. People must also manage several thought pipelines, recover context quickly, judge the quality of the AI’s work and absorb externally generated results back into their own cognitive structure.

As you use Agentic AI, watch your own behaviour. How many thought pipelines are open at once? Can you truly re-enter a problem after switching? Have you absorbed the AI’s output? Did the increase in ability happen inside you, or only inside the tool’s interface?

I do not yet have final answers. But these questions may deserve longer attention than "How much work can AI do for me?"