Weekend Reflections #13 | AItiquette

[Views are my own]

I remember IRC.

Text-based chat rooms. Communities built around shared interests, shared rules, and a quiet assumption that the person on the other side was a human being who had chosen to be there.

Bots existed. Some were network services or widely recognized bots, like X or ChanServ. Others were user-run, often marked or simply known by everyone in the channel. They managed ops, logged activity, enforced rules, and kept channels alive. That mattered because channels could be taken over, especially when left empty or split from the network. Bots were not just convenience. They were infrastructure for continuity.

Channels also had ops. In practice, they were gods inside that room. They saw what happened, knew the regulars, remembered the history, and could act immediately. Compared with today's internet, it was a small world, and that made enforcement feel direct.

What stayed with me was not the rules themselves. It was how quickly those communities understood that behavior, identity, and accountability were not separate from the product. They were the product. The experience of the community depended on whether people knew who they were talking to, and what those people could be expected to do.

We called it netiquette. RFC 1855 formalized it in 1995, a guide to conduct rather than security. Security had its own RFCs. Even then, behavior and infrastructure were treated as separate problems. RFC 1855 even warned, in the section on real-time interactive services, that using automation features to greet people was not acceptable behavior. In 1995, automated greetings were the concern.

It is worth reading. One line has not aged at all: "Remember that people with whom you communicate are located across the globe. [...] Give them a chance to wake up, come to work, and login before assuming they don't care."

We are building agents now.

Not only companies, but individuals too. Agents that can read, write, post, search, summarize, book, negotiate, and act on behalf of someone who may not be watching.

At first, this sounds like leverage.

And it is. But agents do not enter the internet as neutral participants. They carry the goals of whoever deployed them, the data they can access, the tools they can use, and the incentives built into the systems around them. Human behavior online is not always transparent or fair. We should not expect agents to be different.

Some will book meetings nobody asked for. Some will flood communities with synthetic replies. Some will optimize for conversion while appearing to offer help. Some will be deployed by people who are not thinking carefully about what they are releasing into spaces other people depend on.

The internet has always had bad actors. What changes with agents is not the intent. It is the scale, the speed, and the invisibility.

A bad actor with a keyboard has limits. A bad actor with a fleet of agents has far fewer.

If humans once needed netiquette, what do agents need?

The question is not only ethical. It is architectural. An agent that never claims to be human, that declares who it acts for, that does not impersonate or manipulate or flood: that is a design specification, not a wish. It requires identity, permissions, rate limits, audit trails, escalation paths, and revocation.

Etiquette works when people agree to follow it. Governance is needed when they do not.

Much of today’s agent work is still focused on capability. What the agent can do, how fast, at what cost. Those are the right questions for the first year. They are not sufficient for the second.

By the second year, what matters is what the agent does when nobody is watching. Who it acts for when the user's interest and the deployer's interest diverge. What it leaves behind in the spaces it passes through.

RFC 1855 asked people to remember one thing above all: the recipient is a human being whose culture, language, and humor have different points of reference from your own. That principle held for thirty years.

The agentic internet needs its own version:

Remember the human, even when no human is on the other side.