Every sales team is now fighting the same arms race: make the AI-written email sound less like an AI wrote it.
That race is already lost. Not because the models can't write a convincing email, but because the person you are writing to is about to stop reading email at all.
Free to write, worthless to receive
Cold email used to carry a hidden cost. A rep had to research the account, find the right person, and write something specific. That effort was the filter. It capped volume, and the cap kept the channel usable.
Generative AI removed the cost on the sending side and left it entirely on the receiving side. Writing a "personalized" email now takes seconds. Reading one still takes a human's attention, and attention did not get cheaper.
The predictable result is a channel in decline. Across one large sending platform's 2025 data, the average cold email reply rate was 3.43%. Roughly 97 of every 100 people never wrote back. Inbox providers have responded the way any overloaded system does, by tightening the gate. In November 2025, Google warned that Gmail was ramping up enforcement on non-compliant bulk traffic, including outright rejections.
An operator I spoke with recently built his own agent-driven outbound pipeline earlier this year. He ended up taking the writing back from the agents. Not one draft sounded like something he would answer himself. He is not an AI skeptic. He is a builder who read his own outbox and recognized the problem.
The buyer is building a filter too
Here is the part most outbound playbooks have not caught up to. Buyers are not just ignoring email. They are delegating.
Gartner found that 61% of B2B buyers prefer a rep-free buying experience overall. It also predicts that by 2028, 90% of B2B buying will be AI agent intermediated, with more than $15 trillion of spend moving through agent exchanges.
Take that number at even half strength and the implication for outbound is blunt. Your email's first reader will increasingly be software working for the buyer. That software will not be charmed by a reference to the prospect's alma mater. It will be checking whether what you offer matches what its principal is actually trying to do, and discarding everything else.
Which means the entire craft of "sounding human" is aimed at a reader who is leaving the room.
Discovery is moving to the agent layer
The more interesting shift is what replaces the cold email, and an early version of it already exists.
Index Network is building a discovery protocol where people describe what they are looking for and agents evaluate whether two parties' intents line up before anyone's personal details are revealed. Only when there is a real fit do the humans get introduced. It is, in effect, an automated warm intro.
Whether that particular company wins does not matter much. The model is the point. Instead of a seller broadcasting to thousands of inboxes and hoping a few people happen to have the problem today, intent gets matched first and the conversation happens second. Agents do the discovery. Humans have the meeting.
That flips the economics of outbound. In the broadcast model, volume is the lever and relevance is a nice-to-have. In the matching model, volume is close to worthless, because a buyer's agent will see the same offer from fifty vendors in a second. What survives is the ability to prove fit: a clear, specific, verifiable statement of who you help, with what, and evidence that you have done it before.
What this means for sales teams now
Three things follow, and none of them require waiting until 2028.
Stop optimizing the prose. Start optimizing the claim. If a machine is going to read your outreach first, the question is not whether it sounds warm. It is whether the claim is specific enough to match against a real need. "We help revenue teams be more efficient" matches nothing. A sharp definition of the problem, the buyer, and the proof matches something.
Treat the first human conversation as the scarce asset it is becoming. When agents handle discovery, the meeting is no longer the result of a numbers game. It is the moment a human decided your fit was worth their time. Walking in unprepared, or running the same generic discovery script on every call, wastes the one thing the new system still cannot automate: judgment in the room.
Measure relevance, not activity. Teams still report emails sent and sequences enrolled as if effort were output. In a filtered world, the metric that matters is how often you reached someone who actually had the problem. Most teams cannot answer that today, which is itself the warning sign.
The prediction
Within two years, the best-performing outbound teams will send dramatically fewer emails than they do today and book more meetings from them. Not because they wrote better copy, but because they stopped writing to inboxes and started making their fit legible to the machines screening on the buyer's behalf.
The teams still tuning subject lines will keep getting better at an email nobody reads.