A Metamorphosis

I.

When Sam McGregor woke that Tuesday morning, he found that he had, at some point in the night, become a dinosaur.

He did not notice immediately. His first thought, as it was every morning, was the 9 AM stakeholder sync. He reached for his phone to check the agenda and discovered that his arm was now a forelimb: short, scaled, and ending in two small claws that could not unlock the screen. He tried his face. Face ID did not recognize him. He attempted the passcode again. The claws were surprisingly dexterous for their size, but the phone was designed for human fingers, and after three failed attempts it locked him out.

The stakeholder sync would begin in forty minutes.

He sat up. His tail knocked the bedside lamp to the floor. The room looked different from this height, or perhaps from these eyes, which were set on either side of his head in a way that made focusing on anything directly in front of him difficult. He was perhaps ten feet long from snout to tail. His skin was a mottled gray-green and felt, when he twisted to examine it, like a very expensive briefcase.

His first thought was that this would affect his commute. His second thought was that the slide deck for the stakeholder sync had not been finalized, and the revised projections from finance still needed to be incorporated. His third thought was that he should probably tell someone. He was not sure whom. He lived alone in a studio apartment in SOMA, twelve hundred square feet of polished concrete and floor-to-ceiling windows, optimized for a person who was rarely home. He had moved in three years ago and had never hung anything on the walls. There was no one to tell.

He tried to open his laptop. The claws worked poorly on the trackpad. He could not type. He managed, with some difficulty, to nudge the cursor to the Slack icon and to click, though clicking was harder than it seemed because the pressure he applied was no longer calibrated for the resistance of a MacBook trackpad. The app opened. He tried to type a message to his manager. He tried to type anything at all. The claws produced only strings of adjacent keys, incomprehensible.

The stakeholder sync would begin in thirty-one minutes.

He stood. This took some effort, as his center of gravity had shifted and his legs, though powerful, were not designed for standing still. He moved toward the door. Walking was surprisingly natural. He had, it seemed, retained some procedural knowledge even as his declarative memory remained intact. He could walk. He could not type. He could think in user stories but could not write them.

The Slack notification sound came from his laptop. He returned to it, nudged the trackpad. A message from his manager: Running behind today, can you kick off the sync and share the deck?

He tried again to type. The claws produced: qwsdertghj

He deleted it. He tried once more. aszxcfghj

He deleted it again. He clicked on the video icon. The camera opened and he saw himself for the first time: a therapod, something between a velociraptor and a very small tyrannosaur, its scales faintly iridescent under the ring light he used for calls. Its eyes, his eyes, were yellow with vertical pupils. His teeth, visible when he opened his mouth experimentally, were numerous and sharp.

He closed the camera.

The stakeholder sync would begin in twenty-three minutes.

He managed, eventually, to join the call. He did not turn on his camera. He could not speak; his larynx had changed in some way that made human speech impossible, though he found he could produce a range of clicks and low rumbling sounds that might have been language to another dinosaur. There were no other dinosaurs in the product org. He listened as his manager ran through the deck he had built, the projections he had not yet revised, the roadmap he had sequenced. No one mentioned his silence. The stakeholder from marketing asked about the Q3 launch timeline. His manager said they were on track. Sam tried to interject that they were not on track, they had not been on track since the API dependency had slipped, and no one had updated the assumptions, but what came out was a series of clicks that sounded, through his laptop microphone, like static. Someone said, "Sam, you're breaking up." He muted himself.

After the call, he received a Slack message from his manager: Everything okay? You were pretty quiet.

He tried to respond. He could not.

Another message: FYI I'm getting some heat from leadership on velocity. Can you pull together a deck on how we're using the new AI tools? They want to see adoption metrics.

He tried to respond. He could not.

Another: Also quick heads up, we're doing a reorg announcement at 2 PM. Nothing major, just some team shifts. Will send calendar invite.

II.

The accommodation happened faster than he expected.

By Wednesday, someone had moved a standing desk into the large conference room on the fourth floor. The room had been used for all-hands meetings but those had moved to Zoom during the pandemic and never returned. It was empty most days. Now it had a standing desk, a large monitor mounted at his head height, which was now somewhat higher than human head height, and a small placard on the door that said SAM McGREGOR, SENIOR PRODUCT MANAGER.

He did not know who had made these arrangements. He suspected his manager, though his manager had not come to see him since the transformation. Communication had become difficult. He could read Slack messages and emails, but he could not respond to them except by nudging emoji reactions, which he did frequently and which his colleagues interpreted, generously, as a contribution. 👍meant yes. 👀 meant he was reviewing. 🦖 meant he had nothing to add. He used it often.

He attended meetings by video. He left his camera off. Nobody bothered to ask why. The AI notetakers didn’t notice.

The junior PMs had taken over most of his active projects. This was not announced formally; it simply happened. Jordan, who had been on the team for eight months and who had previously asked Sam for advice on stakeholder management and prioritization frameworks, was now running the checkout optimization initiative. She was using Cursor to generate PRDs directly from customer support tickets. She was using Claude to write user stories. She was using an AI agent that someone had configured to auto-create Jira tickets from Slack threads, assign story points based on keywords, and suggest sprint allocations. She was shipping features faster than Sam had ever shipped features.

Sam watched the PRDs appear in Notion. They were well-formatted. They included all the standard sections: problem statement, success metrics, user stories, scope, open questions. They were generated in minutes. They were, in Sam’s estimation, approximately 70% correct, which is to say, they correctly described what the customer had asked for but did not correctly describe what the customer needed, because the customer did not always know what they needed, and because the difference between the two required judgment that could not be extracted from support tickets.

He tried to flag this in a comment. He could not type. He reacted with 👀.

The feature shipped. The metrics looked good for two weeks. Then the support tickets started coming in: not about the original problem, which had been solved, but about a new problem that the solution had created, because the solution had optimized for one user flow while breaking another, and no one had noticed because no one had talked to the users who used the other flow, because talking to users took time and the AI could synthesize support tickets in seconds.

The team did a retro. The AI agent generated a summary of the retro. The summary noted that velocity had increased 40% quarter-over-quarter and recommended continuing current practices.

Sam attended the retro. He could not speak. He reacted with 🦖.

The reorg, when it came, was gentle in its language.

His manager explained it to him over Zoom, camera off, as if matching his state might make the conversation easier. "We're shifting to a leaner model," she said. "More agentic workflows, fewer individual contributors. Your role is being absorbed." She paused. "Absorbed isn't the right word. Redistributed. The work you were doing is being redistributed."

Sam tried to ask where he would be redistributed to. The clicks he produced were interpreted as connection issues.

"You'll stay on through the transition," his manager continued. "We want to make sure there's continuity. Knowledge transfer. You've got context that the AI can't replicate." She paused again. "Yet. Can't replicate yet. But that's why we want you to document everything. Get it into Notion so the agents can access it."

He was being asked to train his replacement. His replacement was not a person. His replacement was a configuration of tools that Jordan had set up in an afternoon and that could generate PRDs and user stories and sprint plans at a rate that made his years of accumulated judgment look like latency.

He tried to explain that the judgment was the point. That knowing what not to build was harder than knowing what to build. That the customers who complained loudest were not always the customers who mattered most. That the features that tested well in isolation often failed in combination. That product management was not documentation and prioritization and stakeholder alignment, or rather that it was those things, but that those things were outputs of a process that happened in the space between meetings, in the questions you asked before you wrote anything down, in the pattern recognition that let you see the third-order effects that would only become visible six months from now when the thing you shipped had reshaped the system it was shipped into.

He could not explain any of this. He produced a series of clicks. His manager said, "I think we're losing the connection."

III.

By the fourth week, the accommodation had become indistinguishable from neglect.

The conference room was still his. The standing desk remained. The monitor displayed his calendar, which was increasingly empty. He was no longer invited to the stakeholder syncs. He was no longer tagged in the PRD reviews. He was cc'd on announcements and weekly summaries, but these were automated, sent by systems that did not know or care that he could not respond.

He spent his days reading. He could still scroll. He could still click. He read the Slack channels, the Notion pages, the Jira boards. He watched the features ship. He watched the metrics dashboards. He noticed patterns that no one else seemed to notice.

The velocity metrics were up. Way up. PRs were merging faster than ever. Features were launching every week. The dashboards were green. Leadership was pleased. The AI adoption metrics that his manager had asked for, weeks ago, before everything, showed that the team was using Claude for 80% of documentation, Cursor for 60% of implementation planning, and the agentic workflow tools for nearly all ticket management. The numbers were presented in a slide deck at the all-hands. The slide deck was generated by AI. The all-hands was recorded and summarized by AI. The summary was distributed automatically to stakeholders who did not attend.

But the customers were not happier. This was the thing Sam noticed. The metrics that mattered — adoption, retention, expansion revenue — were flat. Or rather, they were fluctuating: up one month, down the next, in a pattern that suggested volatility rather than growth. The features were shipping but the needle was not moving. The team was producing more and achieving the same.

He tried to write this down. He spent an afternoon nudging his claws across a large touchscreen that someone had brought him as a gesture of accommodation, or perhaps of guilt. He managed to produce a document. It was short. It was mostly typos. But it contained the core observation: velocity had increased, outcomes had not.

He shared it in the product leadership Slack channel. He could share things, even if he could not write them coherently. The document appeared. No one responded.

Three days later, Jordan shipped a dashboard that showed the same data. She had pointed Claude at the Slack channels and asked it to surface anomalies. Claude had found Sam's document, ingested it, and produced a cleaner version with better formatting and a summary paragraph at the top. Jordan presented it at the leadership review. She was praised for her analytical instincts. The dashboard was added to the weekly metrics email. The weekly metrics email was sent to 340 people. No one mentioned Sam's document. It was possible no one had opened it. It was possible the AI had been the only one to read it at all.

The end came on a Thursday, at 4:47 PM, thirteen minutes before a meeting that Sam no longer attended.

Jordan found him in the conference room. She knocked first, which was polite, though the door was glass and she could see him through it, standing at the monitor, scrolling through a customer research repository that no one had accessed in months.

"Hey," she said. She did not seem uncomfortable. The younger PMs had adapted faster than the older ones, or perhaps they had simply never known him as anything other than the dinosaur in the conference room. "I wanted to give you a heads up. The transition is done. The agentic workflows are handling everything now. The knowledge transfer is complete." She paused. "Well, as complete as it's going to get. The AI has ingested all the Notion docs. It's answering questions about the product better than I can, honestly."

Sam looked at her. He did not know what expression his face was making. He did not know if dinosaur faces could make expressions.

"Anyway," she said. "I just wanted to say thanks. For everything you built. The frameworks, the templates, the processes. The AI uses them constantly. It's like you're still here, kind of. In the system."

She smiled. It was a real smile, not unkind. She meant it as a compliment.

Sam watched her leave. He turned back to the monitor. The customer research repository was still open. He had been reading transcripts from eighteen months ago, when he had talked to users himself, when he had sat in their offices and watched them struggle with the product and asked questions that were not in any template. He remembered a user named David who had explained, over forty-five minutes, why the feature the team was about to build would not solve his problem. Sam had gone back and killed the feature. It had been the right call. The data had not supported it; the data had supported building the feature. But the data had been wrong, because the data measured what users did, not why they did it, and the why had mattered.

The AI could not have made that call. The AI had not been there. The AI had not seen David's face when he described the workaround he had built, the resignation in his voice when he explained that he had stopped expecting the product to help him. The AI could summarize a transcript but it could not feel the weight of that resignation, could not recognize it as a signal that the product had failed in a way that no metric would capture.

But the AI was here now, and Sam was not. Or rather, Sam was here, but Sam was a dinosaur in a conference room, and the AI was in every Slack channel and Notion page and Jira board, answering questions and generating documents and shipping features, and the features were shipping faster than ever, and the outcomes were exactly the same.

He left that evening. Not dramatically; there was no confrontation, no farewell, no final stakeholder sync. He simply walked out of the conference room, down the hallway, into the elevator, through the lobby. The security guard nodded at him. He walked through the parking garage, past the cars he could no longer drive, and out onto the street.

The sun was setting. The air was warm. He walked without destination, his claws clicking on the sidewalk, his tail swaying for balance. People moved out of his way, then continued walking, faces in phones, earbuds in. A child pointed. The child's parent pulled them along. No one stopped. No one called the police. He was strange but not, apparently, strange enough to interrupt the flow of the city.

He walked until he reached the waterfront. He stood at the railing and looked out at the bay. The water was flat and gray. A ferry moved slowly toward the far shore. The sky was turning orange at the edges.

He thought about the roadmap he had built last quarter. It had been good. It had balanced short-term wins against long-term bets. It had accounted for dependencies and risks. It had told a story about where the product was going and why. The AI had ingested it. The AI was now optimizing against it, shipping features that aligned with its categories and metrics, without understanding that the categories had been provisional and the metrics had been proxies for something harder to measure.

The product would be fine. The team would be fine. Features would ship. Dashboards would stay green. The AI would get better at generating PRDs and the junior PMs would get better at prompting it and the whole system would run faster and smoother than it ever had when Sam was part of it.

The outcomes would be exactly the same.

He watched the ferry reach the far shore. He turned and walked north, away from the office, away from the waterfront, away from the city. He did not know where he was going. He did not think it mattered. The system would continue without him. It was continuing without him already. It had been continuing without him for weeks.

Somewhere behind him, a calendar notification chimed on a phone he could no longer unlock, reminding him of a meeting he would not attend, for a team that no longer needed him, on a product that would ship far faster than it had when he was there.

The velocity metrics, he knew, would be up. The outcomes would be flat. Next quarter, someone would conclude they needed to ship even faster.