They Cloned Me, Then Begged

I spent eight years at the company, working my way up from a green rookie to the most expensive expert on the entire East Coast. Then the company gave me an AI assistant. I thought it was a perk. It wrote my weekly reports, answered my emails, and organized my data. I loved using it. Three months later, HR called me in with a termination agreement. "The AI assistant has already distilled you into an AI clone. The company doesn't need you anymore." So that assistant hadn't been helping me. It had been replacing me. I signed and left. Eight years of my life, traded for a cardboard box and two words: "Not needed." Then my old boss called, his voice shaking as he begged. "Meridian's system crashed! The AI gave us seventeen fixes, and every one was wrong. They'll pull the contract if it's not fixed tonight. That's $450 million! Please come back. I'm begging you!" I gripped the phone and heard myself laugh. "It learned every document, every log, every action I ever took. But it'll never learn the one thing that's mine alone: my brain. "When all the data points to A, I have the nerve to pick B. "As for you? Go ahead and keep picking A." Chapter 1 "Miles, the AI's better than you, so we don't need you anymore." The HR rep slid a termination agreement across the table. I froze. I had designed the system architecture for Meridian Group, our most valuable client. Nearly $450 million in annual transactions passed through systems I'd built and maintained. I was the only person at Vantex Systems who understood all their peculiarities: which processes could withstand a beating and which had to be routed around. How could I possibly be worse than an AI? Outside the meeting room, my coworkers were talking just loudly enough for me to hear. "This is hilarious. Miles finally got knocked off his pedestal." "The highest-paid expert on the East Coast, huh? AI showed up and he was the first to go." I curled my hands into fists until my nails dug into my palms. Why hadn't the company trained an AI replacement for them? Because what they did all day—sucking up to management, trading favors, working their connections—really was something AI couldn't learn. Ha. But the future would be driven by technology. People like that wouldn't last. The door opened. Patrick Devereux, our director of technology, stepped inside carrying a tablet. "Miles, this isn't the company deciding to let you go," he said. "The data made the decision." He turned the tablet toward me. A comparison chart filled the screen. Over the previous three months, I'd completed forty-seven support tickets, averaging 3.2 hours each, with one unresolved bug. The AI had completed 142 tickets, averaging four minutes apiece, with zero reported bugs. "Look at it yourself," Patrick said. "The numbers don't lie." I stared at the bug listed against my name. It was an obscure boundary condition I'd discovered during a live incident. I hadn't had time for a permanent fix, so I'd written a fallback to protect production until I could return to it. The AI had never even come near that boundary condition, because all it ever handled were routine tasks. Patrick swiped to the next slide. "Here's the architecture optimization our AI generated last week. It's ninety-two percent structurally similar to the proposal you submitted three months ago, except the AI produced it in three seconds." He paused, as if announcing a scientific finding. "Your irreplaceability has been reverse-engineered, Miles." I knew of three more hidden boundary conditions in the Meridian system. None had ever made it into the documentation. They hadn't occurred in the historical training data, and the company's new digital version of me had no way of anticipating them. But I didn't explain any of it. Even if I did, they wouldn't understand. They'd just think I was making excuses. I signed the agreement. My hand trembled as I wrote my name. Patrick leaned against the doorframe, talking as if he were stating a law of nature. "Don't take it personally. AI replacing human workers is inevitable. You're just part of the first wave. Eventually, everyone will have to accept it." I packed my desk into a cardboard box. Coworkers watched me walk down the corridor the way they'd watch an obsolete server being wheeled out for disposal. Patrick called after me. "Miles! If our AI ever runs into trouble, we'll still know who to call for advice." He sounded polite, but every word cut into me. The elevator doors closed. I leaned back against the wall and shut my eyes. Eight years. Three-in-the-morning alerts. Thirty-six-hour stretches at client sites. The champagne they'd opened when I became the company's senior technical expert. All of it, thrown away for nothing. I opened my eyes. You'll regret this, I thought. Chapter 2 I carried the box home and found my wife, Claire, trying to soothe our crying son, Noah. She glanced at what I was holding, but she didn't ask. Our baby had been born prematurely. His lungs were weak, and even his cries were quieter than other babies'. She'd only just given birth and should've been resting in bed, yet she was the one taking care of both of us. "The hospital called again," she said gently. "We owe another four hundred and fifty dollars." She laid the baby down and went into the kitchen to warm my dinner. She'd gotten painfully thin. I gripped the termination agreement. My throat tightened. My phone rang. A recruiter offered me a chief technology officer position at $300,000 a year. I thought about it for three seconds and declined. A second call offered $270,000; a third offered $225,000. I turned them all down. There was no point in going. They wouldn't see my value either. They'd only put me through the same thing all over again. I was waiting for the right moment. I rented a windowless basement room for $175 a month and told Claire that the company had arranged an apartment for me. She believed me. The moment I shut the door, I told myself, This is your last chance. Before leaving, I'd retained backups of my own notes and the operational data the company had copied from my personal computer: activity logs, decision patterns, and keystroke records. I used those materials to reverse-engineer the AI clone's decision-making architecture. What I discovered was a fatal weakness. The clone relied entirely on historical data. It made its decisions by matching patterns. The moment it faced a black-swan scenario, something that had never happened before, it would fall apart. It couldn't reason. All it had were statistical probabilities. I began building an AI framework of my own. I called it Plan B. Instead of copying past decisions, the framework would test the assumptions behind them. When the available data was incomplete, it would work through alternative explanations and derive a response to circumstances that had never appeared in an operations log. Two weeks later, the hospital called. Our son needed surgery within the next month. We had to come up with almost $30,000. My bank balance was just over $7,000. I ended the call and kept typing. When there's no going back, the only way is forward. At three in the morning, Plan B was ninety-nine percent complete. I pushed the keyboard aside and stumbled across a video of Vantex's latest launch event. Gordon Huxley, the CEO, stood onstage beneath an enormous screen. "A technology-driven management revolution," the screen proclaimed. "No more dependence on supposedly irreplaceable individuals." Patrick stood behind Gordon, grinning. My old coworkers cheered from the audience. "Gordon and Patrick rule!" "The age of AI is here!" I closed the video and went back to work. Let them celebrate. A month after my dismissal, Plan B reached one hundred percent. The company's clone had learned my code, notes, logs, and recorded actions, but it still hadn't learned my mind. When every historical example pointed toward A, I could recognize the moment B was the only sensible answer. They would pay for their devotion to A. I only had to wait. Meanwhile, Vantex went public. Efficiency skyrocketed, and the company raked in money hand over fist. Over the holiday weekend, my social feed filled with photographs from a company trip to Miami. Patrick had taken his team to the beach. One former coworker posted a group photo captioned, [Thanks to AI and Patrick, we finally have lives outside the office.] Then I saw a post tagging me directly. It was from Tyler Boone, a junior engineer I'd personally trained. [Apparently our "irreplaceable" expert is still unemployed, while the rest of us are already financially independent. The future doesn't knock before it replaces you.] Patrick liked the post. On his own page, Patrick posed beside the AI server racks. [The AI era rewards people who know how to use tools. Some people cling to outdated methods, get replaced, and still imagine they're heroes.] My knuckles whitened around the phone. I threw it onto the bed, paced the tiny room three times, and sat down again. I couldn't answer them now. An angry reply would only give them another joke to share. I had to wait for it all to fall apart on its own. Eli Moreno, the only coworker who'd remained a genuine friend, called to check on me. "I'm going to watch them destroy themselves," I told him. He thought I was venting. Three days later, I found a public technical post from Meridian Group discussing its upcoming migration to a new architecture. Buried in the post was a reference to differences between the old and new data-encoding formats. My pulse quickened. I'd found that exact problem three years earlier and submitted a redesign proposal. Patrick had rejected it. Eli called again, this time in secret. "Miles, they're moving Meridian's entire production environment to the new architecture in one week. We've never done anything remotely like this. I have a terrible feeling." "Stay away from the migration," I told him. "Patrick had the AI generate the plan. He's guaranteed it will work. Gordon's over at Meridian right now bragging that we have the most advanced enterprise AI on the planet." I sneered. They had no idea how big that problem really was. I had barely hung up when Eli's messages began flooding in. [IT'S DOWN. Meridian's core system keeps crashing at random. Their production lines have stopped. The AI has proposed seventeen fixes—rollbacks, restarts, parameter changes, replacement modules. They've tried every one. Nothing works, or it breaks something else.] At eight the next morning, sixteen hours remained before Meridian's deadline. Patrick's name appeared on my screen. I let it ring five times before answering. His voice was as condescending as ever. "Miles. It's me." "Patrick. What can I do for you?" "We've run into a minor technical issue. We need you back to help." "I'm not your employee anymore." Patrick started shouting at me. "You left so abruptly that you never completed the handover. We could sue you for that!" They'd ordered me out in a single afternoon and told me I was unnecessary. Now they needed me, and he was still threatening me. I hung up without another word. Patrick called again and again. I ignored him until Eli's number appeared. When I answered, Patrick was the one speaking. "Miles, I was wrong about what happened before..." I cut him off. "When you fired me, did you stop to ask how I would feed my family?" He had no answer. I ended the call. An hour later, Eli texted: [Meridian just cut the deadline to twelve hours. They're also threatening a $75 million breach-of-contract penalty.] I replied with one word: [Wait.] At three that afternoon, Patrick called again. This time, his voice was breaking. "Miles, I'm begging you. Tell me what you want. Anything." I leaned back in my chair. "Three conditions. First, a public apology. Second, $1.2 million. Third, bring the AI server into the lobby and smash it in front of everyone." For a full minute, he said nothing. Then he forced the words out through clenched teeth. "...I'll smash it." I smiled. "I'll be there in two hours." Chapter 3 Exactly two hours later, I walked into the Vantex lobby. The entire company seemed to have gathered to watch. Eli stood in a corner. He gave me a small nod. Patrick waited beside the AI server. His face was dark with anger. A heavy sledgehammer rested on the floor nearby. Gordon Huxley, the CEO, came over all smiles. "Miles, you've been treated unfairly," he began. I walked straight past him. It was just good cop, bad cop. Patrick picked up a microphone. His voice was stiff. "On behalf of the company, I formally apologize to Miles Keating. Our decision to replace him with an AI system was wrong, and we failed to respect his contribution as a technical expert." The whole lobby erupted. I listened without changing my expression. Then Patrick took the sledgehammer and swung it into the server. The impact rang across the room. A drive casing split; a fan assembly flew loose. With each blow, his face grew paler. I watched calmly. It wasn't pleasure. It was long-overdue justice. I went to the operations console. "Everyone out. Give me two hours." Patrick opened his mouth. I turned and looked at him, and for once he decided against offering an opinion. Alone in the room, I examined Meridian's failure logs. All seventeen of the AI's proposed solutions had chased the same mistaken assumption. Roll back. Restart. Change a parameter. Replace a module. Each attempt kept returning to the fault it was supposed to fix. The root cause was exactly where I'd expected it: a hidden constraint in the new architecture was interacting asymmetrically with the old system's data-encoding format. The clone's training material contained no example of this conflict, so it had no reason to challenge the assumption underlying all its fixes. I ignored every solution it had suggested. I opened a new file and wrote a conversion function from scratch. Rather than repairing the incompatibility directly, I designed an asynchronous layer that would decouple the two sides and route the conflicting data around the problem. Forty-seven lines of code. When I typed the final character, I murmured, "The data points to A. The answer is B." I ran the function. The error rate plunged from fifteen percent to 0.01 percent. An hour later, Meridian's systems were fully operational again. The total elapsed time was one hour and forty-seven minutes. Everyone was waiting when I emerged. "One hour and forty-seven minutes," Gordon repeated, looking torn between relief and terror. Patrick's lips had lost all their color. "Remember this," I said. "AI will never replace people. Especially not me. When all the data points to A, I have the nerve to choose B." Gordon forced a smile. "Excellent work, Miles. I'll have Finance arrange the $1.2 million as soon as possible." "Three days," I interrupted. "Put it in writing. Don't play games with me." His smile stiffened. Patrick suddenly stepped forward. "What proof do you have that our AI used your intellectual property? You signed an employment agreement. Legally, the company owns its systems!" I laughed. Something about it made everyone else fall quiet. From my bag, I took out a USB drive. "While you were training the digital version of me, you illegally collected data from my private computer. My logs. My decision patterns. Even my personal notes. None of that is company property. It's my personal intellectual property. "Under privacy law, using my personal data to train a commercial AI without my consent gives me grounds to sue." Patrick tried to argue his way out of it. I waved the drive. "There's also an email you sent on here: 'Collect as many of Miles Keating's decision-making characteristics as possible.' Want to take a look, Patrick?"