A software engineer at a 100-person company recently described a pattern that's becoming recognizable in AI-equipped workplaces: he spent 90 minutes writing careful technical feedback on a colleague's project. The reply came back with clean subheadings, addressed each of his points in sequence, and read nothing like how that colleague normally writes.
The colleague had apparently run his comments through an AI tool before responding.
This dynamic - where human effort becomes the raw material for someone else's AI output - is spreading as companies do broad tool rollouts. The company in question had just given everyone access to Cursor, an AI-powered coding environment (think of it as a code editor where an AI assistant can read your entire codebase and write code directly inside it). When an AI assistant is already open in the background, using it to help structure a written reply takes about 30 seconds.
The Cost Is the Signal
There's a real argument for using AI to improve written communication - clearer structure, no burying the lead in paragraph four. That's different from what's happening here. The engineer's feedback required real context: understanding the technical approach, knowing what had been tried before, applying judgment about trade-offs. That thinking produced a document with specific, substantive concerns.
When that gets fed into an AI, the model doesn't evaluate the concerns - it categorizes them, reformats them, and returns a structured acknowledgment. Each point gets addressed at surface level. Nothing gets added. The 90-minute review became a prompt.
What disappears is the signal that someone engaged with the work. A thoughtful pushback means the reviewer saw something the author missed. An AI-generated response that ticks every box is indistinguishable from one that didn't see anything at all.
What Broad AI Rollouts Actually Change
The productivity case for tools like Cursor is well-documented for code work. Less examined is what happens to organizational communication when everyone has the same AI assistant running. People start using it for everything - not because they're lazy, but because it's there and it's fast.
The result is a kind of AI telephone game: you write carefully, the recipient feeds your message to a model, the model generates a reply, they approve it. Both sides use AI to mediate the exchange. The humans in the loop are mostly editing AI output rather than communicating.
This matters beyond individual frustration. Technical decisions made through AI-mediated communication tend to resolve surface disagreements without surfacing the underlying ones. The model can't identify what it doesn't know. So the feedback loop looks productive while the real work of reaching agreement gets skipped.
The practical response: route genuinely important feedback through synchronous conversation rather than documents. A 15-minute call where both parties are actually present beats a well-formatted written exchange where neither is. That's not a knock on AI tools - it's recognizing which problems they're actually good at solving.