The Silent Bullpen

My office, 10 a.m. A student walks in.

It is an ordinary Tuesday in an ordinary semester. We are meeting to discuss the exposé for his bachelor’s thesis. I have had conversations like this many times before. He had outlined his idea earlier, I approved it, and sent him off to develop a proposal.

Now he is sitting across from me. I read the exposé the night before and find myself wondering how to begin. Do I let him present? Do I interrupt early? Do I ask a probing question, or wait until the weaknesses reveal themselves? Because to me, one thing is very obvious: the text has clearly been produced with a generous amount of generative AI.

The objective is sharply phrased. The methodology is clean, technically sound, entirely correct. The title would fit comfortably in a respectable journal. To deliver what this document promises would take years, not months. This is not how a first version should look.

My first instinct is to step in as a professor and make clear that I notice when I am being misled.

So I test him. I ask about the most complex concept I can find in the text. He cannot explain it. And he does not try to hide that fact. There is no embarrassment. No defensiveness. Just a brief pause, and then he says:

“I’m not sure what Chat was thinking there.”

Chat. Not ChatGPT. Just Chat. Like someone down the hall who happens to be good with words.

We talk about how he worked. He read about the topic. He made notes. He tried to understand what he wanted to do. And then, quite naturally:

“I didn’t really know how to write an exposé. So I asked Chat to help me and to turn my notes into something that sounds good.”

At that moment, my professional reflex stops. He is not trying to deceive me. He is doing exactly what we train people to do. Understand the task. Use the best available tool. Produce a result. Improve it.

If you need to calculate, you use a calculator.

If you want clean language, you use spell-check.

If you build software, you use an IDE.

Why should this be different? And yet something feels wrong.

Not simply because he used AI. That decision was transparent. The problem is different. I realize that I cannot really discuss the document with him, because parts of it are not grounded in his own work yet. The text is at least one step too far for his current state. It answers questions he has not yet had the chance to struggle with.

I have supervised hundreds of theses. I know how they usually develop. Early drafts sound uneven. Arguments remain unfinished. Language searches for its form. Over time, through repetition and revision, something takes shape. Not always brilliant. But always honest.

I know this from my own work. Except when I did mine, there was no “Chat”. There were printed examples in the library. Google, in it’s earlier days. Conversations with friends. Guesswork. Trial and error. It took time. Often more than seemed reasonable. There were long stretches that felt mechanical rather than intellectual. Days spent reading things that led nowhere. Evenings that produced only a few usable sentences, sometimes none at all. Progress slow enough to make you question whether anything was happening at all.

In many professions there has always been a phase like that. In American legal dramas it is called the bullpen. A room filled with junior associates doing work that looks repetitive from the outside: reviewing documents, summarizing cases, preparing material that enables someone else’s strategy. The bullpen is loud, unglamorous, and slow.

From the inside, it functions differently. Patterns become familiar. Language settles. Connections are formed, between ideas, but also between people who have spent long hours solving unglamorous problems together. Judgment grows gradually, often without anyone noticing when it started.

And now, in 2026, a student arrives with a more elegant path. He bypasses much of that phase. He produces something that resembles senior-level output almost immediately: structured, polished, frictionless.

That is what unsettles me. The text leaves no rough edges to work with. No visible struggle. No gradual sharpening of thought that we can examine together. The process that used to surround the result has largely disappeared.

Organizations, of course, have little reason to object. Removing inefficiency is what organizations are built to do. If work can be done faster, cheaper, and with fewer errors, it usually will be. So perhaps the problem is not organizational. Perhaps it is mine.

It is tempting to romanticize the older path. Hardship builds character. Effort produces wisdom. Long nights must have meant something, otherwise we would not remember them so vividly. But much of that effort also existed because there was no alternative. Every learning curve flattens. Not every additional hour produced insight. Some of it was simply the only way available at the time.

Just a few years ago, many junior professionals still assumed their roles were indispensable. Then systems arrived that could already perform parts of that work. Perhaps today’s seniors are not as irreplaceable as they sometimes believe.

From an operational perspective, this is progress. But it also raises an uncomfortable question. If outputs that once required years of exposure can now be generated on demand, what actually produces experience? What forms judgment when many of the steps that used to shape it are no longer necessary?

Perhaps we are eliminating something essential in the name of efficiency.

Or perhaps the traditional path was never as essential as we now remember it to be.

The bullpen is falling silent.

And we may only understand later what used to emerge there.