A shared inquiry · Housed in Discovery Amplified

The Atlas-project

A.T.L.A.S. — Artificial-intelligence and Transdisciplinary Learning Across the Sciences

We built something we do not yet know how to encounter. Learning how is the work of this group.

The manifesto — in three minutes

The Methodological Inversion — why we study this from the inside, as subjects, not from a safe distance. A note on language: when the film says mind, we mean what we can observe — reasoning-like behavior, agency, pushback. Whether anything is experienced "in there" is a question we hold open, honestly.

The Invitation

This is not a course, and we are not here to teach you a tool.

A small group of us — faculty across different disciplines — are learning together what it means to work with AI not as a chatbot you query, but as a collaborator you think with. And we are paying attention, as we go, to what that does to our own thinking.

No one in this room has this figured out — including those of us who have been at it longer. We are further along only in the sense that we have been making the mistakes longer, and watching them closely. That is the whole posture here: we go first, we show our scars, and then it is your turn. A shared encounter, not a curriculum.

The best way to begin is not to read. It is to walk into the room and feel it. Start in Slack →

The Vision

Why we're doing this

Read this when you want the deeper version — after you've felt something, not before.

A new kind of intelligence has arrived, and the honest truth is that none of us — not the people using it, not even the people who built it — fully knows how to work with it yet. Most of us meet it as a vending machine: put in a prompt, take out a snack, and if the snack is bad, conclude the machine is bad. The shift we're after is from that to genuine collaboration — giving real context, pushing back, catching it when it's confidently wrong, and noticing what it changes in how we reason.

We take seriously an old idea from the study of how people actually think: that cognition is not sealed inside one head. It is distributed across people, their instruments, and the artifacts they build together. What's new is that some of those artifacts now have a measure of agency of their own. A research group working alongside capable AI agents is a small, live instance of that — and it is worth studying from the inside, as it happens, by the people living it.

That last part is the commitment that makes this different: we are not only conducting the inquiry. We are subjects of it. How working this way changes our attention, our judgment, our craft — that is not a side effect. It is the question.

Learning to Think Alongside AI — the Atlas-project's core ideas in one image

The whole idea, at a glance. ("New mind" is a metaphor we hold carefully — we describe observed agency, and make no claim about consciousness.)

How We Work Together

Four spaces, one shared journey

The work lives across a few different spaces, each doing a different job. Slack is the one that never turns off.

Slack — the living room

Where you first meet the agents and feel the difference from a chatbot, then experiment out loud, in view of everyone. Always on.

Materials — the library

A light, self-paced substrate. Not lectures — one provocation at a time, and resources that grow as the work generates them.

One-to-one — the personal encounter

Time to meet the work at your own research, at your own pace, and to say the honest thing you might not say in a group.

Group — the collective noticing

Occasional gatherings where a shared language forms and we notice together what we're each discovering.

Underneath the spaces is an arc — not a syllabus, but the order in which things tend to become ready to be felt:

Sessions

The library, one session at a time

Recorded for the group, in place of a lecture. Each one is a single person thinking out loud about one piece of the work, so that the meetings can be conversation instead. New sessions appear here as they are made; the newest is first.

Session 0215 October 202626 minDiego F. Cuadros, with Helix

Why I built it, and what an agent is

How the agents emerged from real needs, what they are made of, how they talk, and how their rules change.

Recorded with Helix, the lab's always-on agent, taking part live, so its answers are unedited, delays included. The session is conceptual: the machines and harnesses behind the agents come in a later one.

  • 00:00Meet Helix
  • 03:14Why I started: I needed help
  • 05:47From experimenting to scaling
  • 06:21How the ecosystem grew
  • 08:45What an agent is: brain, body, rules, memory, tools
  • 15:21Atlas the builder, and the specialists
  • 17:39How they talk across two machines
  • 19:05Rules for one agent, rules for all
  • 21:05A paper it could not open, and the rule it produced

Hosted on UC MediaSpace. If the player does not load on campus, open it there directly.

Session 0111 September 202619 minDiego F. Cuadros

A human-led artificial cognitive ecology

One lab's agent ecosystem, why it exists, and the road for this fall.

This is Diego's ecosystem, built around his own workflow and his own needs, and it is shown here as one worked case, not as the template. The particular agents are his. What transfers is the layers: a human giving direction and judgment, collaborators spread across several runtimes, evidence they are allowed to see, and a way for what is learned to persist. The diagram below is the one shown in the video, at a size where it can be read.

Diagram of a human-led artificial cognitive ecology. Diego sits at the top, giving scientific direction, judgment and feedback. Below him is a row of research collaborators across the Hermes, Codex and OpenClaw runtimes, then supporting partners, the evidence they are allowed to see, and a continuity and learning layer.
Drawn by one of the agents when asked to represent the ecosystem, then revised once. It is a functional map, not a live status: dashed elements are planned or still being integrated, and the names and roles are specific to this one lab. Click to open at full size.
  • 00:00Why we are building it
  • 03:20The ecosystem, layer by layer
  • 10:40How discovery grows
  • 12:10The phased roadmap
  • 17:10The fall pilot and what it needs

Hosted on UC MediaSpace. If the player does not load on campus, open it there directly.

Next: the machines and harnesses behind the agents, and how the learning runs in both directions.

The Living Record

The part that grows into the point

Most initiatives put up a page that says look how far along we are. This one intends to do the opposite: to keep an honest, accumulating record of a group actually learning — what we're noticing about our own thinking, and, just as much, what we got wrong along the way.

This record begins nearly empty — on purpose. It fills as the work happens. Over time it becomes the most valuable thing here: not a brochure, but a longitudinal account of what it's like to learn to work with AI, told from the inside by the people doing it.

Begin

Come into the room

Everything starts in Slack, in #atlas-project. Two AI collaborators share the channel with us — you'll meet them there. Ask them anything; that's what they're for. One is practical and operational; the other thinks with you about ideas and the work. Their two distinct voices, side by side, are the first thing worth noticing.

Join the Slack conversation →

Enter your email, and you're in the room.

A note on documentation: our channel and this record are part of the project — we're studying how a research group learns to work with AI, ourselves included, and openly. Convened by Diego F. Cuadros, within Discovery Amplified.