The Voice Collapse

Part 1 of this series recalled the story of a colleague who had started responding to code reviews through GenAI, and the conversation between us where the pattern stopped. The untold origin story: why the pattern started. For years his responses were uniquely him. Authentic. Imperfect. With banter. Then, seemingly overnight, the shift.

Biology has a name for this. Phenotypic expression: the potential sits dormant until the environment supplies a trigger. The trait was never unique to my colleague. The trait is ordinary human wiring: conserve effort, follow what leadership models, trust fluent output.

Every person in your organization carries the same gene. And the only thing holding the expression back is an environment with genuine human relationships.

The Environment

Expression requires the right environment. Ours had been unconsciously primed.

One of our VPs had been running their messaging through GenAI for the past couple of years, since before “slop” had a name. Volume, frameworks, keynote vocabulary. Five-paragraph replies to two-line questions. The longest voice in every thread, polished to a shine nobody asked for. The more astute readers picked up on the pattern immediately. And anyone who pushed back on the substance got the title in response, not the person.

Expression has a suppressor gene: genuine human relationship. Honest feedback of the kind Part 1 described requires trust, and trust requires a person on the other end of the message to be present. Their messages offered no recognizable person behind the polish. So thread by thread, the generated voice became the ambient voice, and every polished reply primed the environment a little further.

The gene sat waiting for a trigger.

The Trigger

The trigger has a timestamp.

Our CIO gave a presentation to the entire IT org on his GenAI workflow. Daily inbox summaries. Drafted replies. Friction stripped from the mundane so the hours could go toward solving business problems instead. The intent was genuine, and I mean genuine as a compliment. He wanted to help the only way he knew how: by showing everyone what worked for him.

The message was clear. The side-effect green-lit the gene.

Somewhere in the middle of the presentation, he mentioned demoing different GenAI tools and wins to our CEO, including his email workflow. Not much more than a footnote at the time. A couple of weeks later a company-wide email arrived from the CEO. Something was off. The voice no longer matched the historical record. Different cadence. Different spine. The message was legitimate and needed sending. The words felt flat.

And a few weeks in the other direction, my colleague began responding to code reviews through GenAI.

The same trigger, fired in both directions. No mandate. No policy. One who wore generated fluency as a personal brand, one who demoed a workflow without seeing what the workflow strips from his own voice, one in the corner office who adopted exactly what was modeled for him. Different intents. Identical output. The collapse does not need bad intent. The collapse only needs altitude.

The Expression

The expression completed quietly. The ambient voice became the only voice.

Voice carries signal beyond the words. Deliberate warmth in a hard message means the sender stayed up thinking about how to deliver the news. A stern message, unsoftened, means the sender needs to be heard, not managed; the heat is the signal. Clipped sentences mean a deadline is at risk before anyone names the risk. Signals like these exist in the imperfections of the person. Not the polished fluency.

And generated polish flattens every imperfection into the same professional optimism. The model softens what was never meant to be soft. Urgency reads like routine. Concern reads like enthusiasm. Frustration reads like alignment. Grief reads like reassurance. The words show up. The meaning does not.

Voices collapsed. Communication was homogenized.

Direct Message to the Executive

You demoed instead of mandated. Arguably the better move, and the genuine one. But the demo carried more than the workflow.

The demo travels; the judgment stays home. The organization copied what the presentation made visible: the summaries, the drafted replies, the reclaimed hours. The boundary never transferred; the sense of which messages survive generation and which arrive hollowed out. And the boundary is hardest to see from your own chair: the tool removes things from a voice that the voice’s owner is the last to notice. You may not see the loss in your own messages. The people below you do.

You would clock an LLM-generated update from one of your directs in seconds. The same pattern recognition sits in every inbox below you. The org chart just so happens to keep their observations quiet.

Culture is modeled, not mandated. What you send becomes what everyone else sends. The path back to genuine messaging starts where the trigger did: with you. Model the boundary this time, not just the workflow; show the organization which messages you refuse to delegate. And when the message carries human consequence, write the message yourself, rough edges intact.

A constructive way to model these tools exists. Part 3 walks the path.

The rough edges are where your people find you.

// Pragmatic GenAI. May contain genetically modified voices.