Pilot Purgatory - Part 4
Pilot Purgatory
A Story of AI Transformation
By Scott Weiner (AI Lead at NeuEon, Inc.), inspired by conversations with Erwann Couesbot (CEO of FlipThrough.ai)
Note: This is a work of fiction. All characters, companies, and events are fictional composites created for illustrative purposes. While the industry statistics cited are real and sourced, the narrative is designed to illuminate common patterns in enterprise AI adoption, not to depict any actual organization or individuals.
This is Part 4 of a serialized story exploring why enterprise AI initiatives fail, not from lack of technology or talent, but from invisible organizational dynamics that doom them from the start.
Reading the series for the first time? Start with Part 1: The Mandate
Previously in Pilot Purgatory…
Linda Chen faced 32 vendor proposals with vocabulary she couldn’t verify. “Enterprise-ready.” “Seamless integration.” “State-of-the-art.” Every proposal used the same buzzwords, but none could demonstrate their platform working with Thornfield’s actual data.
The demos were polished, and revealing. Priya Gupta asked vendors to test against real data extracts. They couldn’t. “We typically do that during the pilot phase, after the contract is signed.”
Meanwhile, Priya’s data audit uncovered an uncomfortable truth: 43% of Thornfield’s product specifications had conflicts, 18% were missing entirely, and 7% were demonstrably wrong. Her recommendation, six months to standardize the data foundation, was met with “Do what you can with what we have.”
The $800,000 Quartzvane contract was signed on a Tuesday. Professional services arrived with laptops and optimism. By week one, their expressions of professional neutrality had shifted to concern. “Your data architecture is more complex than we anticipated.”
The timeline was already slipping. The data gaps waited to become cracks…
Chapter 4: Departure
The LinkedIn notification appeared at 7:43 AM, sandwiched between a connection request from someone Daniel had never met and an endorsement for “Python” from someone who had never seen his code.
“Hi Daniel! I came across your profile and was immediately impressed by your background in ML/AI implementation. Our client, a cutting-edge AI-first company, is seeking talented engineers like yourself to join their mission of revolutionizing the enterprise space with next-generation artificial intelligence solutions. Would you be open to a quick chat about an exciting opportunity that aligns perfectly with your career trajectory?”
Daniel read the message twice. The phrasing was too smooth, too optimized for keywords, too perfectly calibrated to hit every box on a recruiter’s engagement checklist. It read like it had been written by exactly the kind of AI he spent his days building.
He almost deleted it. The inbox was full of messages like this, each promising exciting opportunities and competitive compensation and remote-first cultures. Most led nowhere. Some led to companies that wanted AI skills but had no AI work. A few led to conversations that ended with salary offers that felt like insults wrapped in enthusiasm.
But something made him pause. Maybe it was the seven months of accumulated frustration. Maybe it was the third legacy system he had debugged this week, the one built on a framework that had been deprecated before he finished college. Maybe it was the feeling of watching his AI skills sharpen at Thornfield while knowing the company couldn’t afford to let him use them full-time.
He typed a response: “I’m open to learning more. What’s the company?”
The reply came within the hour. A real conversation this time, not automated responses. The company was Terfim AI, a Series B startup in the Bay Area that was building tools for enterprise customers. Fully remote. Pure AI work. A team of thirty engineers who did nothing but machine learning.
The recruiter asked if Daniel was available for a call that week.
He looked around the engineering floor. Half the monitors showed Quartzvane dashboards, the vendor platform that was still being integrated three months after contract signing. The other half showed maintenance queues, the endless tide of support tickets and bug fixes that kept the legacy systems running.
He opened his calendar and found a slot.
The call happened on Thursday, during what Daniel had blocked off as “focus time” that no one else respected.
He took it from a coffee shop three blocks from the office, sitting in a corner booth with his laptop open and his earbuds in. The recruiter was professional, knowledgeable, and refreshingly honest about both the opportunity and the challenges.
“The role is pure AI engineering,” she said. “No legacy maintenance, no context switching, no splitting your time between initiatives. Our engineers spend their days building models, not keeping old systems alive.”
“What’s the compensation?”
She named a number. Daniel set down his coffee.
“That’s forty percent more than I’m making now.”
“We pay market rate for AI talent. The market rate has shifted significantly in the past eighteen months.” A pause. “I understand if you need time to process. But I should mention that we’re moving quickly on this role. The team wants to make a decision within the next two weeks.”
Daniel thought about his desk at Thornfield. The fine-tuned models he had built. The chatbot that was finally handling customer queries with reasonable accuracy. The predictive maintenance analysis that was still waiting for clean data that never seemed to arrive.
He thought about Marcus, who had given him the AI projects in the first place. Who had believed in him before the skills existed. Who had carved out space in the budget for experimentation when other priorities were screaming for attention.
“I’ll think about it,” he said.
“That’s all I’m asking.”
He hung up and stared at his cold coffee. The numbers echoed. Forty percent raise. Pure AI work. No legacy maintenance.
The math was simple. The decision was not.
Daniel didn’t sleep well that weekend.
He ran the calculations over and over. His current salary versus the offer. The cost of living adjustment for remote work versus Boston area. The value of Thornfield’s equity, whatever that was worth after fifteen years of steady but unspectacular growth.
The numbers always came out the same way. The offer was better, significantly better, the kind of better that would affect his savings rate, his timeline for buying a house, his ability to help his parents when they needed it.
But the numbers didn’t capture everything.
They didn’t capture Marcus’s face when Daniel had first shown him the fine-tuned model, the genuine excitement of a CTO who was learning to see AI as something more than a checkbox on a board presentation. They didn’t capture the late nights when Marcus had stayed to understand, asking questions that revealed curiosity rather than just obligation.
They didn’t capture the team he had helped build, the junior engineers who came to him with questions about embeddings and attention mechanisms, the data analysts starting to understand why AI needed clean foundations, the customer service managers finally seeing how automation could help rather than threaten.
And they didn’t capture the guilt. The feeling that leaving now, when the project was halfway through its most difficult phase, would be abandonment dressed up as career advancement.
He thought about what his father would say. His father who had worked at the same manufacturing plant for thirty-seven years, who valued loyalty the way other people valued promotions. Who would see the job change as a betrayal even though he would never use that word.
He thought about what his career counselor would say. The one from business school who had told him that loyalty to companies was a relic of a different economy, that smart professionals moved every two to three years, that staying too long in one place was a sign of stagnation rather than commitment.
Both perspectives made sense. Neither felt complete.
On Sunday night, he opened his laptop and typed a message to the recruiter: “I’d like to proceed to the interview stage.”
The response came in minutes: “Wonderful. I’ll send calendar invites for the technical and culture rounds.”
Daniel closed the laptop and lay in the dark, trying to convince himself he was making the right choice.
The interviews happened over the following week.
The process included a technical round testing his knowledge of transformer architectures and fine-tuning methodologies, a culture round exploring his collaboration style and communication preferences, and a final conversation with the hiring manager, a former Google engineer who asked thoughtful questions about his experience at Thornfield.
“What’s been the biggest challenge?” the hiring manager asked.
Daniel considered the question. He could have talked about the data quality issues, the vendor integration delays, or the security concerns that kept getting pushed to the backlog.
Instead, he told the truth.
“Time fragmentation. I’m doing AI work maybe ten, fifteen hours a week. The rest is legacy maintenance, support tickets, keeping old systems running. The AI projects are interesting, but they’re not my job. They’re something I squeeze in around my actual job.”
The hiring manager laughed, not unkindly. “Let me guess. You’re also on-call for production issues. And when something breaks at 2 AM, nobody cares that you were supposed to be training models.”
“Last month I spent three days debugging a payment gateway integration. Zero AI work that week.”
“Yeah.” The hiring manager leaned back. “Here, if you’re debugging a payment gateway, something has gone seriously wrong with your job description. You’d be doing AI work. That’s it. That’s the whole job.”
Daniel felt something shift in his chest. The weight of fragmentation lifting, replaced by the possibility of focus.
“That sounds like what I’ve been looking for.”
“Then let’s make it happen. I’ll have an offer to you by end of week.”
The offer arrived on Friday afternoon.
Daniel read it three times. The base salary was higher than the recruiter had mentioned. The equity package was meaningful. The benefits were comprehensive. The start date was flexible.
Everything about it was better than what he had.
He printed the offer and stared at it for a long time, the Thornfield logo visible through his office window where it was mounted on the atrium wall. He had walked past that logo every day for four years, and it had meant something once: stability, opportunity, the kind of company that invested in its people.
It still meant something, just not enough to offset what he was being offered elsewhere.
He scheduled a meeting with Marcus.
Marcus’s office was quiet when Daniel arrived. The afternoon sun slanted through the windows, casting long shadows across the desk where the stress ball sat, momentarily still.
Daniel set his Thornfield mug on the edge of the desk. He had bought it at the company store during his first week, back when such purchases felt like investments in belonging. Now it felt like a goodbye.
“I wanted to tell you in person,” Daniel said.
Marcus looked at the mug, then at Daniel, his expression shifting through surprise, processing, and finally resignation. “The offer. What is it?”
“Forty percent raise. Fully remote. Pure AI work, no legacy maintenance.” Daniel sat in the chair beside Marcus, not across from him. Closer. The way you sat with someone you respected, not someone you were confronting. “I tried to make this work.”
“I know.” Marcus’s voice was quiet. The stress ball found its way into his hand. “I know you did.”
“You gave me the AI projects. You fought for the budget. You let me learn on company time when other managers would have said I should be fixing tickets.”
“And now someone else is offering to pay for what you learned.”
“That’s not how I see it.”
“But it’s what’s happening.” Marcus squeezed the stress ball. The logo on its surface had been worn away years ago. “What would it take to keep you? I can go to David. Twenty percent at least. Maybe dedicated AI time, a formal arrangement.”
Daniel shook his head, something sadder than refusal in the gesture. “It’s not just the money. It’s the split. I’m doing AI for ten hours a week and spending thirty maintaining systems built before I graduated high school. Even if you got me dedicated time, how long would it last? Until the next fire? Until the next board presentation needs a demo?”
“If we got you fully dedicated, “
“For how long?” No anger in Daniel’s voice, which somehow made it worse. “The structure doesn’t support it. You’re stretched across a dozen priorities. The team is stretched. The budget is stretched. AI needs full-time attention from people who have full-time AI jobs. That’s not what Thornfield can offer.”
Marcus set down the stress ball. Through the window, the engineering floor was visible, half-empty in the late Friday afternoon. “I learned everything about transformers on company time,” Daniel continued. “You gave me that. I won’t pretend otherwise. But I can’t stay in a holding pattern forever.”
“I know.” Marcus did know, that was the worst part. “When’s your last day?”
“Two weeks. I’ll make sure everything is documented. The models, the configurations, the lessons learned.” Daniel paused. “I’m sorry it has to be like this.”
“So am I.” Marcus stood, extended his hand. The handshake was firm, professional, and somehow final. “You’re making the right choice. I hate saying that, but you are.”
Daniel nodded and left the office, the mug still sitting on Marcus’s desk where he had placed it.
The two weeks passed in a blur of documentation and knowledge transfer as Daniel wrote guides for the models he had built, recorded videos explaining the fine-tuning process, and sat with junior engineers walking them through codebases and configurations, trying to compress four years of accumulated knowledge into transferable form.
It was never going to be enough, he knew that. The knowledge that lived in his head, the intuitions about what worked and what didn’t, the pattern recognition that came from debugging the same systems for years, none of it could be fully captured in documentation.
But he tried.
On his last day, he packed his desk in a cardboard box that was smaller than he expected, holding a few personal items, some reference books, and a stack of company t-shirts accumulated from hackathons and team events.
One shirt caught his eye: “Thornfield Innovation Sprint 2022.” He remembered that hackathon, three days of intense work producing a prototype that had impressed the judges and been greenlit for development before being quietly shelved when other priorities consumed the engineering resources. The project had never been deployed; the shirt was all that remained.
He placed it in the box with the others and carried everything to his car. The parking lot was half-empty, the way it always was on Friday afternoons. The building’s glass facade caught the setting sun, its surfaces clean and modern, giving no indication of the complexity that lay inside.
He drove away without looking back.
Jennifer Park’s call came three weeks later.
Marcus took it in his office, the door closed, the stress ball already in his hand. Board members didn’t call CTOs casually. There was always an agenda, even when the conversation started with pleasantries.
“I wanted to check in on the AI initiatives,” Jennifer said. “The board meeting is in six weeks, and I’m trying to get a sense of where we stand.”
“We’re making progress.” Marcus heard the words come out and recognized them for what they were: the platitude he used when the truth was too complicated for a quick conversation. “The Quartzvane integration is moving forward. The chatbot is handling a growing percentage of customer queries. The procurement forecasting model is in testing.”
“What about the team? I heard you lost an engineer.”
“Daniel Park. He took an offer we couldn’t match.” Marcus paused, choosing his next words carefully. “It’s part of the market. AI talent is expensive, and the demand is outpacing supply.”
“Are you concerned about continuity? Losing institutional knowledge?”
“We’re managing it. Daniel documented everything before he left. The team is cross-trained on the critical systems.”
Jennifer was silent for a moment, the kind of silence that suggested she was deciding whether to push harder or accept the answer.
“I’m sure you have it handled,” she said finally. “But I want you to know I’m watching this closely. The board has significant investment in these initiatives. We need to see returns.”
“You will.”
“I hope so, Marcus. I hope so.”
The call ended. Marcus set down the phone and stared at the ceiling.
Jennifer was watching. The board was watching. Everyone was watching to see if the AI investment would pay off. And the team that was supposed to deliver that payoff was shrinking while the expectations kept growing.
He thought about Daniel’s words. The structure doesn’t support it. AI needs full-time attention from people who have full-time AI jobs.
He pushed the thought away. There were tickets to review, meetings to attend, a business to run. The AI work would happen alongside everything else, the way it always had.
He hoped it would be enough.
The second departure came eight weeks later.
Maria Chen, the machine learning engineer who had joined six months after Daniel, handed in her notice on a Tuesday morning. The offer was from a competitor in the autonomous vehicle space. The raise was thirty-five percent.
“I’m sorry,” she said, the words echoing in Marcus’s office the way Daniel’s had. “The work here is interesting, but I need to be somewhere that’s all-in on AI. Not somewhere that’s fitting it in around other priorities.”
Marcus made a counteroffer. Maria declined politely.
The third departure came three weeks after that.
James Liu, the data scientist who had been building the procurement forecasting models, accepted a position at a healthcare startup. The raise was forty percent. The role was dedicated ML work, no split focus, no legacy maintenance.
“You trained me well,” James said on his last day. “Better than you know. That’s why someone else is willing to pay more.”
Marcus nodded. There was nothing to say that he hadn’t already said twice before.
On the night after James’s departure, Marcus sat in his home office with a spreadsheet open on his laptop. The coffee beside him had gone cold, untouched, the surface developing the skin that came from hours of neglect.
He was calculating the cost.
Not the cost of departures. He knew that number. Three engineers gone. Three salaries saved, at least temporarily. Three positions to fill, eventually, at rates that would be higher than before.
He was calculating the other cost. The invisible one.
Time to competency was six months minimum, which was how long it took a new AI engineer to understand Thornfield’s systems, data structures, and specific challenges. His three AI engineers had lasted an average of fourteen months, becoming competent at month six, valuable at month nine, and gone at month fourteen, yielding just eight months of full productivity before the market pulled them away. The cost of training, including salary during ramp-up, senior engineer time for mentoring, mistakes made during learning, and productivity lost to knowledge transfer, added up quickly.
For each engineer, the cost was somewhere between $150,000 and $200,000, invisible on any line item or budget report, but real.
Three departures meant half a million dollars in invisible costs, and every departure took knowledge with it that couldn’t be documented, transferred, or replaced.
He thought about Jennifer’s question. Are you concerned about continuity?
He had said no, that they were managing it.
The spreadsheet told a different story.
The pattern was clear now, visible in the numbers the way it hadn’t been in the conversations. Thornfield was training AI engineers for other companies, investing in skills that would be deployed elsewhere, building a talent pipeline that flowed outward instead of upward.
He typed the words into a blank cell, watching them appear in the spreadsheet’s empty corner: “The Invisible Talent Tax.”
That’s what this was. Not just attrition. Not just market forces. A tax, paid invisibly, continuously, in knowledge and capability that accumulated elsewhere.
Marcus closed the spreadsheet and stared at the dark window. His reflection stared back, a man who hadn’t blinked in minutes, trying to solve a problem that wasn’t really about engineering at all.
The structure doesn’t support it.
Daniel’s words again. The truth that Marcus had known but hadn’t wanted to name.
AI needed full-time attention from people who had full-time AI jobs. Thornfield couldn’t offer that, not really, not when every AI engineer was also a legacy maintainer, a support ticket responder, a context switcher between the future and the past.
The competitors understood this, Axiom understood it, and the companies hiring his engineers understood it. They built structures that supported focus, created roles that allowed depth, and paid premium rates because they could extract premium value.
Thornfield was paying to train people for those companies. Every dollar invested in AI skills was, partially, an investment in someone else’s workforce.
He thought about the board meeting in five weeks, the progress report he would have to give, and the questions Jennifer would ask.
He would talk about milestones achieved and metrics improved, present the roadmap and timeline and resource plan, be honest about challenges and optimistic about solutions.
But somewhere underneath the presentation, the invisible tax would continue to accumulate. The cost of pilot purgatory that no budget report could capture.
He closed his laptop and sat in the darkness, trying to figure out how to solve a problem that kept getting more expensive the longer he avoided naming it.
To be continued…
What happens next: The chatbot finally goes live, and for two glorious weeks, everything works. Customer satisfaction scores tick upward. Response times plummet. Then Gerald Patterson, a maintenance supervisor in Ohio, asks a simple question about warranty coverage. The answer the chatbot gives is confident, detailed, and completely wrong. Chapter 5 reveals how a single hallucination can unravel months of progress and expose every shortcut taken along the way.
Part 5 publishes January 28, 2026.
Why we wrote this
Scott Weiner is the AI Lead at NeuEon, Inc., where he helps organizations navigate the complexities of AI adoption and digital transformation. This story draws from patterns observed across dozens of enterprise AI initiatives.
Erwann Couesbot is the CEO of FlipThrough.ai, specializing in AI strategy for professional services. His conversations with technology leaders inspired many of the dynamics explored in this narrative.
Reading the series for the first time? Start with Part 1: The Mandate
Missed Part 2? Read Chapter 2: Foundations
Missed Part 3? Read Chapter 3: Procurement
Want to read the complete story?
Have your own AI transformation story? We’d love to hear it. Connect with Scott on LinkedIn or reach out to NeuEon at neueon.com/contact.