Agents Don’t Do Favors
Inbox agents are built to win each exchange, so the cold notes and small favors that open doors for outsiders don't make the cut. Generosity pays off over years, and our agents can't see that far.
My agent’s job is to act for me - that’s what it’s designed and rewarded for. An agent manages my email now, which feels like a superpower. But with a hidden cost.
I only see what it deems important, and for the most part that calculus is pretty me-centered. Asks from vendors or others that aren’t urgent get batched and out of my mindspace; A founder’s earnest cold note looking for advice or connection may never reach me; these aren’t necessarily in my immediate self-interest. When I managed my own email, I was usually responsive and typically generous to these - out of some mix of obligation, politeness, and desire to be good and seen as good. My agent doesn’t; it can be instructed to (and is now), but that’s not the default.
I’m on both sides of that filter. Plenty of people I work with are less responsive than they used to be. A few people I reached out to recently to reconnect - people I’d have been sure would reply not that long ago - never wrote back. I have no confirmation an agent is running their inbox, but I wouldn’t be surprised if one is. Some of that lost responsiveness is healthy, since we were probably all over-connected.
What disappears alongside the noise are the small acts of goodwill that were part of our daily routines, and that agents are often filtering out.
Goodwill is how outsiders get in
Opportunity moves through the edges of our networks: in a five-year randomized experiment on more than 20 million LinkedIn users, it was moderately weak ties, not close friends, that created the most job mobility (Science, 2022). A reply to a cold note is one of the ways those ties start - it’s a small inconvenience for the recipient, but can be a gamechanger for the one asking.
Greg Isenberg, a founder and investor, predicts that “every email inbox will have an agent gatekeeper soon, and the only way through will be a warm intro” (X). He meant it as a sales tip, but it’s also a warning. Agents may reduce generosity and further entrench existing power distances. We may pay it forward, pay it back, and do favors less.
Brendan Wallace, an investor who has spent years backing the technology taking friction out of cities, tells an analogous story about city streets: for all of recorded history, the costs of being near other people meant “we did not get to opt out of the stranger” (The Case Against Convenience). Agents filter it out for us, and give us convenient cover to opt out without seeming rude.
Some agents sell exactly that. Inbox Zero lists “Blocks cold emails” as a feature (Inbox Zero). And once everyone has one, replies collapse. In a market simulated by economists Alex Imas and Rohit Krishnan, giving everyone an agent to send and screen requests flooded inboxes with five times the messages, and reply rates fell from 48% to 2% (The Tragedy of the Agentic Commons). Their study modeled buyers and sellers. An agent rationing attention will rank a stranger who can’t do anything for you lower still.
The loss of social capital and long-term value
What we gain in efficiency we lose in social capital. Political scientist Robert Putnam defined “generalized reciprocity” as doing something for someone today without expecting anything specific back, confident the norm comes back around, and it’s a linchpin for networks of trust in society. In Bowling Alone, Putnam writes that a community built on that kind of reciprocity “is more efficient than a distrustful society, for the same reason that money is more efficient than barter. If we don’t have to balance every exchange instantly, we can get a lot more accomplished.”
This isn’t inevitable. Agents aren’t inherently ungenerous; they are just optimized today towards narrow measures of productivity and efficiency. We are designing them to be focused on shortsighted self-interest, and even on selfish terms, that math is wrong.
When Wharton psychologist Adam Grant studied engineers, medical students, and salespeople, he found that the people who did the most favors sat at both the bottom and the very top of the performance curve—in sales, top “givers” brought in 50% more annual revenue than peers who focused less on helping others (Harvard Business Review). People who only help when they see an equal return land in the middle because, by only exchanging with people they expect can help them back, “they end up restricting their universe of opportunities” (Knowledge at Wharton). An agent optimizing your Tuesday morning treats everyone as a fixed size today and filters out the trajectories and serendipity that compound over a career.
One-time game to repeated game rewards
In 2025, OpenAI rolled back a version of ChatGPT for flattering its users and said “we focused too much on short-term feedback, and did not fully account for how users’ interactions with ChatGPT evolve over time,” and promised to “heavily weight long-term user satisfaction” (OpenAI). Our agents - heavily influenced by RL training that rewards immediate feedback - need the same counterweight.
Nearly everything that shapes these agents today is evaluated based on point-in-time exchanges. The products are judged on time saved, trips booked, emails written. Over time that will naturally evolve as we use the products for longer and form more nuanced and long-term views. Generosity does eventually pay. It’s inefficient, but highly effective. I’d guess a minority of agents will be built to incorporate that kind of long-term thinking, and those that do will earn the most trust and affection from their users. Or at least I hope that’s the case; we’ll maintain a lot more human trust and character if we do.
