Why Those Who Build Technology Are Most Scep

The server room hummed with the quiet persistence of machines that had no intention of failing, nor any particular ambition to succeed. In the corner, two figures – Thomas, a backend engineer approaching his fifteenth year with the company, and Marcia, newly appointed as Head of AI Strategy – spoke in the careful tones of people who suspected they were discussing entirely different subjects.

Backstage at the Revolution

The executives had gone home hours ago, leaving behind glossy presentations with confident predictions about machine learning transforming their logistics chain. The deadline was impossible, the expectations unreasonable, and no one who understood the system had been consulted beforehand.

Thomas, who had not been invited to the meeting, skimmed through the slides Marcia had brought him. He possessed the peculiar stillness of the habitually disappointed.

“They want this implemented across all ten million endpoints by March,” said Marcia. She was American and still believed in the possibility of miracles, especially those announced in boardrooms.

“Ten million,” he repeated, without adding that it had taken eighteen months to deploy the last system update to just fifty thousand nodes. He had learned years ago that some knowledge was not welcome, particularly knowledge that reflected poorly on the vision of progress.

“The demo went perfectly,” she added, with a hint of accusation. She was wearing earrings that caught the fluorescent light – small, tasteful diamonds that suggested she operated in a world where appearances mattered.

“Demos generally do,” Thomas replied, turning to the final slide which promised ‘seamless integration’ and ‘94% accuracy across all use cases.’

The consultants who had produced these figures had never attempted to deploy anything across the company’s sprawling, contradictory infrastructure. They had not encountered the undocumented dependencies, the silent failures, the institutional resistance that manifested as apparently innocent technical questions during implementation calls.

“I suppose you’ll tell me it can’t be done,” said Marcia, who had already committed to quarterly milestones.

“It can be done,” Thomas said, “but not by March, not across ten million endpoints, and not with 94% accuracy.”

Marcia looked at him with the particular intensity of someone recalculating their position. “The board is very excited about this initiative.”

“The board is always excited about initiatives,” Thomas said. “They’re less excited about implementation details.”

In the corridor outside, the cleaning staff moved with the efficiency of people who knew exactly what they were doing and had reasonable expectations about the results. Thomas envied them.

The Temple of Impossible Knowledge

The company’s central office occupied what had once been a Victorian bank. The executive floor retained ornate ceilings and imposing wooden doors, suggesting permanence and authority. The technology department operated from the basement, which had once housed the vaults. The symbolism was not lost on anyone.

Davis, the Director of Enterprise Architecture, maintained an office that straddled both worlds – it had the high ceilings of authority but was tucked away on a half-landing that executives rarely visited. His desk held a precise arrangement of objects: a scale model of the company’s server farm, a Newton’s cradle that nobody was permitted to set in motion, and three identical blue binders with no labels.

“You’ve seen the presentation, then,” he said to Thomas, who had come directly from his meeting with Marcia. Davis had worked for the company for twenty-seven years and had the confident serenity of a man who had survived twelve CTOs and nine complete transformations of corporate IT strategy.

“I’ve seen it,” Thomas confirmed.

“And?”

“And it’s exactly what you’d expect from McKinley Consulting. Big on vision, light on implementation detail.”

Davis nodded, touching the Newton’s cradle without setting it in motion. “They don’t understand the knowledge problem.”

The knowledge problem was Davis’s particular obsession. He maintained that certain types of knowledge could not be encoded for machines because they existed only in the habits and experiences of the humans who possessed them. He had been making this argument since the early days of expert systems in the 1990s, refining it with each new wave of AI enthusiasm.

“They think it’s just a matter of scale,” Davis continued. “As if ten million is just fifty thousand multiplied by two hundred. They don’t understand that scale changes the nature of the problem.”

Thomas was familiar with the argument but let Davis continue. The older man’s perspective had been vindicated too many times to dismiss.

“The board wants to implement the logistics prediction system across all territories simultaneously. They’re convinced it will save eighteen percent on delivery costs.”

“And will it?” Thomas asked.

“It might save five percent in the territories where we have clean data and standardized processes. In the others, it will create expensive confusion.” Davis straightened one of the blue binders. “But that’s not what they want to hear.”

Marcia appeared at the door, slightly out of breath. “I’ve been looking for you both. The CEO wants an implementation plan by tomorrow morning.”

“An honest one?” asked Davis.

Marcia hesitated. “A realistic one.”

“Those aren’t always the same thing,” Davis said, finally setting the Newton’s cradle in motion. The steel balls clicked against each other with the precise rhythm of consequences following actions.

The Vertical Specialist

Eleanor’s office was not, strictly speaking, an office. It was a converted supply closet on the engineering floor that she had claimed during a period of organizational confusion following a merger. The walls were covered with diagrams of the company’s pharmaceutical supply chain, annotated in a cramped handwriting that only Eleanor could decipher completely.

She had been with the company for nine years and was responsible for the inventory management system used in the European pharmacies division. Her system worked so well that executives rarely mentioned it, focusing instead on more troubled areas of the business. This suited Eleanor perfectly.

“I hear they’re planning to replace your system with the new AI solution,” Thomas said, standing awkwardly in the doorway. There wasn’t room for two people in Eleanor’s office unless one of them was unusually small.

“They’ve been planning to replace my system for four years,” Eleanor replied without looking up from her screen. “So far they’ve spent eighteen million pounds on three failed attempts.”

“This one’s different, apparently. It uses machine learning to predict inventory requirements.”

“My system already does that.”

“Yes, but yours isn’t AI.”

Eleanor turned to face him. “My system uses regression analysis based on five years of sales data, adjusted for seasonality, local health trends, and regional demographics. It’s accurate to within four percent.”

“But it doesn’t use neural networks,” Thomas said, aware of the absurdity even as he voiced it.

“It doesn’t need to. Neural networks are for problems where you don’t understand the underlying patterns. I understand pharmacy inventory perfectly well.”

This was undoubtedly true. Eleanor had worked as a pharmacist for twelve years before retraining as a software engineer. She knew which medications were frequently prescribed together, which ones experienced seasonal demand, and which ones sat on shelves until just past their expiration date.

“The board wants a system that works across all divisions, not just European pharmacies.”

“Then they should hire people who understand the other divisions as well as I understand pharmacies,” Eleanor said. “Artificial intelligence is a poor substitute for actual intelligence.”

The Implementation Gap

The meeting room was officially called Strategy Suite 3, but was universally known as the Disappointment Chamber. It was here that ambitious plans came to terms with practical realities. The room’s rectangular table was just slightly too small for the number of chairs arranged around it, ensuring that at least two people would spend any meeting with their backs uncomfortably against the wall.

Marcia had assembled what she called a “tiger team” to deliver the AI implementation. Thomas was there, as was Davis. Eleanor had been invited but had sent a detailed email explaining why she couldn’t attend, along with a twelve-page analysis of why the proposed system would fail in the pharmacy division.

The consultant from McKinley, Gabriel, stood at the head of the table. He was younger than Thomas had expected, perhaps thirty, with the particular confidence of someone who had never had to implement his own recommendations.

“I understand there are some concerns about the timeline,” Gabriel began, glancing at the slides on his tablet. “But our analysis shows that with the right approach, we can absolutely meet the March deadline.”

“Does your analysis include the integration requirements for the legacy systems in our Asian operations?” Davis asked.

Gabriel hesitated. “We’ve allowed for general integration complexity in our estimates.”

“But you haven’t actually examined the systems.”

“We’ve reviewed the documentation.”

Davis smiled thinly. “The documentation is incomplete. The developers who built those systems left during the 2018 restructuring.”

“We can reverse engineer the requirements,” Gabriel said with the optimism of someone proposing to decipher hieroglyphics over a weekend.

“Perhaps,” Davis conceded, “but not by March.”

Gabriel looked to Marcia for support. She was reviewing Eleanor’s email with increasing concern.

“Have you seen this analysis from the pharmacy team?” she asked.

“We’ve reviewed similar feedback from several departments,” Gabriel acknowledged. “But it’s natural for specialized teams to be protective of their existing solutions. They don’t always see the bigger picture.”

“Eleanor sees the bigger picture perfectly well,” Thomas said. “She just doesn’t think your solution will work for her division. And she’s probably right.”

Gabriel straightened his posture. “Our system achieved 90% accuracy in the test environment.”

“The test environment doesn’t have ten million endpoints operating across nineteen regulatory frameworks with forty-seven different inventory management approaches,” Davis pointed out. “Scale changes the nature of the problem.”

The room fell silent except for the soft hum of the air conditioning. Marcia looked at the faces around the table, at the consultant whose confidence was beginning to show cracks, at the engineers whose expressions suggested the resigned patience of people who had witnessed this scene many times before.

“What would you recommend?” she asked Davis, her tone suggesting she already knew the answer wouldn’t align with the board’s expectations.

“Start small. Identify one division where the AI approach actually solves a real problem. Implement it there, learn from the experience, then gradually expand.”

“That could take years,” Gabriel protested.

“Yes,” Davis agreed. “Complex things often do.”

The Human Quantum Advantage

The canteen on the fourth floor had originally been designed to encourage spontaneous collaboration between departments. In practice, it was where people went to complain about meetings. Thomas found Marcia there the day after she had presented the revised implementation plan to the board.

“How did they take it?” he asked, setting down his cup of tea. The canteen served coffee from an elaborate Italian machine, but the tea came in bags from a supplier whose primary qualification appeared to be affordability.

“Better than I expected,” Marcia admitted. “They weren’t happy about the timeline, but Davis made a compelling case.”

Thomas nodded. “He’s been making that case for thirty years.”

“He said something interesting after the meeting. He called it the ‘human quantum advantage.’”

“Ah, his latest theory. That human cognition operates in quantum problem spaces that silicon-based computing can’t replicate.”

“Do you believe that?”

Thomas considered the question. “I believe there are types of knowledge that resist digitization. Not because we lack the technology, but because of their nature. They’re embedded in context, in experience, in the unspoken habits of people who’ve been solving problems for decades.”

“Like Eleanor and her pharmacy system.”

“Exactly. She doesn’t just know the patterns in the data. She knows which patterns matter and why. She knows when to follow the algorithm’s recommendations and when to override them. That kind of judgment is incredibly difficult to encode.”

Marcia looked at her coffee, which had grown cold. “So what’s the point of artificial intelligence if it can’t capture that kind of knowledge?”

“That’s asking the wrong question,” Thomas said. “The point isn’t to replace human intelligence but to complement it. AI is excellent at finding patterns in vast amounts of data. Humans are excellent at understanding which patterns matter in specific contexts.”

“So we need both.”

“We’ve always needed both. The mistake is thinking one can replace the other.”

Through the canteen window, Thomas could see the company’s fleet of delivery vans in the parking lot. Each had once been driven by a person who knew their route intimately – which streets to avoid at certain hours, which customers preferred their deliveries left with neighbors, which loading docks were particularly difficult to navigate. Now most of the routing was automated, but the drivers still made dozens of small decisions each day that no algorithm had yet captured.

“The board has approved a pilot in the Nordic region,” Marcia said. “If it works there, we’ll expand gradually.”

“It might work,” Thomas allowed. “Especially if we’re realistic about what it can and can’t do.”

“And if it doesn’t?”

“Then we’ll learn something valuable. Which is more than you can say for most failed IT projects.”

Marcia smiled for the first time that day. “You’re not what I expected, Thomas.”

“No?”

“When I was told about the skeptical engineers in the basement, I imagined people who were afraid of new technology. But it’s the opposite, isn’t it? You understand it too well to be dazzled by promises.”

“Understanding the limitations of a technology isn’t skepticism,” Thomas said. “It’s the foundation of using it effectively.”

Outside, a driver was explaining something to a new employee, pointing to different streets on a map spread across the hood of a van. The knowledge was being transferred the way it had been for generations – one person showing another how things actually worked, regardless of what the manual said.

Some things, Thomas reflected, hadn’t changed much since the days of apprenticeships. And perhaps they never would.

The server room continued its quiet hum, indifferent to the human drama it hosted. The machines processed their algorithms with neither ambition nor disappointment, while the humans around them continued the messy, essential work of deciding what really mattered and why.