Continued from Page A6 — “THE INTERNET LIVES!”
The People Who Make It
Inside the Human Side of the Bespoke AI Model Factory: Workers Who Left Other Industries, Found This One, and Won’t Stop Talking About It
By KAREN WIERZBICKI / Tribune Staff Writer
Part III: The Craftsmen and Women
The second floor of the factory is organized around a central corridor, with Calibration Suites on either side. The suites are separated by glass partitions — frosted at the bottom for privacy, clear at the top for the sense of connection, the company says, that makes people work better when they can see that their colleagues are also working. It is not a tech-campus open-plan office, which, as Operations Manager Wierzbicki says with visible relief, “is a thing we specifically designed against.”
In Suite 7, Paula Kowalik — thirty-eight, originally from Waukesha, former radiological technician at a hospital whose orthopedic department closed in 2024 — is reviewing the output logs from a training run that completed overnight. She has been at this for three months, long enough to have strong opinions about the right approach to what she calls the “voice problem.”
“Every model,” she said, without looking up from the screen, “has a voice. You can tune it to speak at the right level, use the right vocabulary, match the client’s own communication style. But the voice has to feel authentic. It has to feel like a real voice, not a simulated one. When you get that wrong —” she paused to make a gesture suggesting catastrophe — “the client knows immediately. They can’t always say what’s wrong. They just say it feels… off. And they’re right. Because it is off.”
“Getting it right,” she said, “is the whole job.”
Paula Kowalik came to this job, she said, because of radiology. “In radiology, you are reading images that tell you something real about a real person’s body. You are looking for truth. The picture doesn’t lie to you. Either there’s something there or there’s not.” She finally looked up. “Here, I’m doing something similar. I’m reading what the model produces and asking: is this true? Is this the thing it’s supposed to be? The picture doesn’t lie to you. Either the model is right or it isn’t.”
Down the corridor, near the east window that overlooks a parking lot that has not yet been filled with cars because today is the first day and the workers drove in at various times and parked with the cheerful randomness of people arriving somewhere for the first time, Model Fitter Bobby Stankiewicz is running through intake protocols that won’t be needed until after the public opening, practicing questions that he says he writes himself.
“The company gives you a framework,” he said. “Questions to ask in each session. A template. But the template is — it’s a starting point. The real questions, the ones that tell you something useful, you have to find those yourself. You have to listen to the client and hear what they’re not saying as much as what they are. Like any conversation.”
He is twenty-seven years old and he reads a great deal, which he did not mention but which was visible in the way he talks and in the fact that when he reached for an analogy, he said the intake process is “a little like Studs Terkel, honestly. You’re just listening. You’re trying to understand a person’s work, their life, what they need. And then you’re using what you learned to make them something better than what they’d have gotten otherwise.”
Bobby Stankiewicz’s father ran Stankiewicz Auto Parts for thirty-two years until the chain parts stores came in and made the independent shop unworkable. “My dad knew every car on the block by sound,” he said. “He could tell you what was wrong with your engine before you got out of the car. The chain stores can’t do that. The database can’t do that. You need someone who’s been listening to engines for thirty years.” He gestured at the door of the intake suite. “That’s what we’re doing here. Learning to listen. Building something that listens.”
“My dad knew every car on the block by sound. The chain stores can’t do that. You need someone who’s been listening for thirty years. That’s what we’re doing here.”
Bobby Stankiewicz, Model Fitter • Bespoke AI Model Factory
Part IV: The Third and Fourth Floors
The third floor is where the Research and Development team lives, in a space that is deliberately not described on the public tour — “for now,” Wierzbicki said, with the specific inflection of a man who has been told to say “for now” — except to say that it contains forty-two people doing work that the company characterizes as “foundational model development,” which is to say: making the bolt of fabric better, so that every suit made from it is better than the last.
“The base model improves continuously,” Wierzbicki said. “Every client interaction teaches us something. Every calibration session produces data that makes the next calibration session more informed. The factory gets smarter as it works.” He paused. “Don’t quote me on ‘the factory gets smarter.’ That sounds either very exciting or very alarming depending on who reads it. Let’s say: we learn from experience. Like craftsmen do.”
The fourth floor is the executive suite, and also the one room in the building that looks, to this reporter’s eye, like a place where actual decisions get made. It is spare and serious and has a very long conference table and a set of windows that look north, over the town, toward the lake, which on this Wednesday afternoon is glinting in the early September light the way lakes glint when the summer is starting to give way to fall and the light has turned from white to gold.
From this window, the Mayor of the Internet stood on Wednesday afternoon, after the tour, and looked at his town. You could see the square from here, and the library whose second floor houses the council chambers, and the steeple of St. Casimir’s, and the parking lot of Henke’s Piggly Wiggly, and, if you looked carefully, the blinking construction lights at the site of the new Civic Center where the Badgers will play hockey.
“You know what that is?” he said, pointing at nothing in particular and everything at once. “That’s a town. That’s a real town. With a grocery store and a city council and a hockey team and a church with a pierogi feed on Saturday.” He turned from the window. “The internet used to feel like that. Like a place where real life happened. We want it to feel like that again.”
He turned back to the window. He looked at his town for a long time.
“We’re going to make it feel like that again,” he said. Not to this reporter. To the window. To the town. To the lake.