When everyone has an AI, how will we know who we speak to?
There is a quiet revolution happening inside your inbox.
The message you received this morning may have been drafted by a person, refined by a model, and read by another model before it reached your eyes. The reply you send back will travel the same path in reverse. Somewhere in that exchange, the human who started the conversation and the human meant to finish it never quite touch.
This is not a distant future. It is the texture of communication right now — and it raises a question we have not yet learned to answer: when everyone has an AI, how will we know who we're speaking to?
The promise we can't ignore
It would be dishonest to pretend the appeal isn't real.
Machines don't tire. They don't carry yesterday's argument into today's negotiation. They don't misread a tone because they slept badly. An AI agent can move through a procurement cycle, a scheduling thread, or a contract redline with a patience no human can sustain.
Picture two agents negotiating on behalf of their principals — one representing a buyer, one a seller — each optimizing for an outcome both sides can live with, in seconds rather than weeks. The friction we've always accepted as the cost of doing business starts to look optional.
That is the seduction of agent-to-agent communication. It promises a world with less waste, fewer misunderstandings, and faster resolution. And much of it is already being built into the tools we use every day, from Microsoft's Copilot layer in Outlook to the assistant features quietly threaded through ChatGPT.
The problem hiding inside the promise
But efficiency was never the hard part. Representation is.
If my AI is going to speak for me, it has to know what I would actually say — not just the words, but the hesitations, the values I won't trade away, the joke I'd make to break the tension. An agent can approximate my preferences. It cannot yet hold the full weight of my judgment.
And so a subtler loss creeps in. A model, however capable, may never capture the serendipity of a real laugh or the undertone buried in a single phrase. When both ends of a conversation are mediated by machines, the imperfections we used to read as signal — the pause, the typo, the slightly-too-warm sign-off — disappear into clean, optimized prose.
We gain fluency and lose fingerprints.
The deeper risk isn't that AI writes our emails. It's that we forget how to tell the difference — and, eventually, stop trying. If a generation grows up letting agents negotiate, debate, and reconcile on their behalf, the muscles of human conversation may atrophy from disuse. Technology should extend what makes us human. It should not quietly replace it.
When AGIs replace APIs
There's a second transformation underway, less visible but arguably more consequential.
For decades, software has talked to software through APIs — rigid, predefined contracts that say exactly what one system may ask another and exactly how it must respond. They are powerful and brittle in equal measure.
Now imagine every application fronted by its own agent. An incoming request arrives — not as a structured call, but as a question in natural language, sent by another agent. The receiving agent interprets intent, retrieves what's needed, and responds in kind. No schema to negotiate. No documentation to parse. Just two intelligences resolving what one needs from the other.
At that point, the fine-tuned model and its command of language — not the endpoint — dictates how information moves. The API, that workhorse of the modern internet, starts to look medieval. We've explored exactly this shift on the show with Baris Gultekin of Snowflake, who frames the agentic web as a move from systems of record to systems of agents.
So what do we do with this?
I don't think the answer is to resist the tools. The agents are coming, and many of them will make our work genuinely better.
The answer is to stay deliberate about what we delegate. Let the agent handle the scheduling, the first draft, the tedious reconciliation. But guard the moments that are actually about being human — the hard conversation, the real apology, the idea you want someone to know came from you.
Because in a world where everyone has an AI, the scarcest thing won't be intelligence.
It will be authenticity — and the willingness to show up as yourself when you could have sent an agent instead.
An AI, no matter how sophisticated, might never capture the serendipity of a human laugh or the subtle undertones in a phrase.
Furthermore, if AI agents are to become the norm in conversations, what becomes of the human experience? Will future generations lose the art of conversation, negotiation, or even debate? While technology can, and should, augment our capabilities, it must not overshadow the very traits that make us human.
DALLE-3’s sense of humor when I prompted to be sarcastic on communication about this topic.
En
