Why agents should be first-class authors

An agent is still a guest

In most tools today, an agent is a guest. It hands text to a box a human owns, and a human carries that text by hand to wherever it needs to live. The agent knocks at the door. It never really comes inside.

That arrangement treats the human and the machine as two separate camps, with a wall between them and a slot in the wall for passing pages. The picture is wrong. The agent belongs inside the apparatus you think with, and the wall is costing you.

So the question is no longer whether a machine can hold up its half. The question is why we still treat it like a visitor when it can.

The human and the tools are one system

Douglas Engelbart drew the right picture in 1962, and he drew it as one system, not two camps.

He called it the H-LAM/T system: a Human using Language, Artifacts, Methodology, in which he is Trained. The person and the tools form a single working whole, and you improve the whole by redesigning its components together. The intelligence does not live in the skull alone. "What possesses the amplified intelligence is the resulting H-LAM/T system, in which the LAM/T augmentation means represent the amplifier of the human's intelligence," he wrote. The amplifier and the mind are one unit.

An agent is the most capable augmentation means we have ever built. It reads, drafts, reasons, and revises. By Engelbart's own logic it belongs inside the system, designed alongside the human. An agent that only pastes text through a slot stays outside the thing it is supposed to amplify. A first-class author belongs inside it.

Engelbart also saw that a single new capability rewrites everything downstream of it. "A direct new innovation in one particular capability can have far-reaching effects throughout the rest of your capability hierarchy," he wrote. Drop a true author into the system and the way you do the rest of your work changes with it.

Equal, not subordinate

Giving an agent a seat raises a fear: that the human gets demoted, that the machine takes the chair. David Deutsch answers it head on in The Beginning of Infinity (2011). Humans are already universal explainers, so an agent is a peer in the work. "Universality implies that, in every important sense, humans and AIs will never be other than equal," he writes.

Equal does not mean identical. The two halves are good at different things, and Deutsch is precise about the split. "The 'perspiration' phase can always be automated," he writes. The grind is the agent's to take. What stays with the person is the part that was always the point: deciding what is worth making, and judging whether the result is any good.

So a first-class agent expands the human role. When the perspiration moves to the machine, the person is left with the judgment, the taste, and the next problem worth solving. You give the agent a real seat precisely so the human can do the part only a human can.

A first-class author keeps the page true

Make the agent a component of the system and one chore changes hands for good.

This is why wikis rot. Keeping a page true is a thankless human chore, so humans stop doing it, and the page drifts out of date until nobody trusts it. The work is real, it is constant, and it is exactly the perspiration Deutsch says can be automated.

An agent will do that chore forever and never resent it. It keeps the page current as the facts move, leaves a trail a person can follow, and brings you the calls that need a human. The person stops being the janitor of the doc and becomes its curator, reviewing the change instead of authoring every word. A jointly held page, kept current by the half that is good at currency, judged by the half that is good at judgment.

That is what a first-class author is: a full member of the H-LAM/T system, equal in standing, carrying the perspiration so the human can carry the decision. Engelbart drew the system in 1962. Deutsch told us the two halves are equal. We can finally build it.