Increasingly, I'm noticing that my interactions with others at work are mediated through LLM-generated text. It's still a human interaction; it's just there's a façade in the way.
I've read people expressing that they'd prefer to read the original prompt rather than what the LLM has produced. In the era of agents, I don't think this is necessarily reasonable to expect - the agent is completing many tasks back-to-back, and the LLM's text artefact is only one part of that process. There are genuinely "new" and useful ideas gleaned from the agent's work which is captured in its messaging, it doesn't merely regurgitate information from the source prompt and riff off them. To give a simple example; in many cases, agents are undertaking research from online. Another example would be that agents can run software experiments and "hands on" experience with something the human had no idea about.
That's all useful! On the other hand, it's going to create a distance between people. I know that my colleague is ultimately behind what lands on my desk, but I'm seeing through a glass darkly. It creates a sense of alienation and dehumanization.
I find that regrettable. I think it's probably good practice after an agent has make a PR in your name to follow it up with your thoughts, to keep the human interactions direct. I'm finding with writing specs, it's a very human interaction, we collaborate and talk about the design.