**Artifact** from Bead: Netlogo User Conference 2026 · [canonical source](https://redfish.acequia.io/guerin/.agents/0a0d4bbc-5310-486b-97af-0e16646347a4/2026-06-20/artifacts/alt-let-the-turtles-loose.md) · session 2026-06-20 · discussion: Talk: Netlogo User Conference 2026
### A Declaration for Agent-Oriented Architecture — restructuring computation beyond simulation, toward a deployed civilization of the collective Mind *Stephen Guerin (Redfish) — carrying the lineage of Owen Densmore.* *In dialogue with Uri Wilensky & Seymour Papert, "Restructurations: Reformulating Knowledge Disciplines through New Representational Forms" (Constructionism 2010). In the ethos of John Perry Barlow.* > Draft for the NetLogo conference. The author line, affiliations, and the biographical specifics in §5 are placeholders for Stephen's confirmation. This is the rhetorically-reposted version produced by the `rhetoric-style` bead; the appeals analysis and the before/after of the abstract are in `rhetorical-review-and-appeals.md`.
## Abstract Wilensky and Papert taught us that the deepest gains in knowledge come not from better ways of *teaching* a subject but from better ways of *representing* it. Change the infrastructure in which a discipline is encoded — Roman numerals to Hindu-Arabic, prose kinematics to algebra, the geometric point to the Logo turtle — and you change what can be known, how easily, and by whom. They called such a change a *restructuration*, and they named agent-based modeling as the restructuration of our scientific age: the turtle lets a child understand a circle as a local act of walking rather than a global definition requiring a center; a thousand simple turtles make statistical mechanics visible to a fourteen-year-old. This paper argues that the next restructuration is already underway, and that it walks out of the classroom. When agents stop *modeling* the world and begin to *run* it — when a turtle is no longer a simulation of an actor but a deployed, addressable, self-sovereign agent in production — agent-based *modeling* becomes agent-oriented *architecture*. The representational infrastructure Wilensky restructured for *learning* becomes the operational infrastructure for *building*. We do not want merely an open-source repository. We want this paradigm taken seriously as design, proven by real-world deployment as a lesson from industry, and recognized as a new ontology: computation not as a model *of* reality but as the structure that *is* the system — least action made executable. We foreground the half-century practice of Owen Densmore, from the pre-commercial dawn of personal computing through a decade of AgentScript, as evidence that this is not a thesis but a life's work. And we set it against the centralizing gravity of contemporary AI: against intelligence *collected*, we declare for intelligence *collective* — a civilization of linked minds that the weary giants of the cloud did not build and cannot own. **Keywords:** agent-oriented architecture, restructuration, AgentScript, NetLogo, agent-based modeling, least action, computation as ontology, decentralization, collective intelligence
## 1. The restructuration that has no home Wilensky and Papert's move is simple and radical at once. The science of learning, they observe, has spent its energy on the *means* of instruction — which method teaches fractions better — while taking the *object* of instruction as fixed. Their inversion: the object is not fixed. The encoding of a discipline in a given representational infrastructure is a *structuration*; a change of infrastructure that re-encodes the discipline is a *restructuration*. The paradigm case is arithmetic: the shift from Roman to Hindu-Arabic numerals did not teach multiplication better, it made multiplication a thing ordinary people could do at all. The dramatic gain came from the representation, not the pedagogy. They then name the cost of this insight: the science of restructurations *has no natural home*. It needs deep disciplinary knowledge, creativity in the design of representations, and sensitivity to learning — and no department, no funding line, is shaped to hold all three at once. They report this as a problem of academic structure. We accept the argument entirely, and we read the homelessness as a symptom of something larger than academia. **A restructuration powerful enough to change what a discipline *is* will not stay inside a curriculum.** It pushes outward — from how we *teach* a field to how we *practice* it, and finally to how we *build the world* with it. The turtle was never going to be content drawing circles on a classroom screen. The homelessness Wilensky names is the restlessness of an idea that belongs in production.
## 2. From the circle to the commons: when the turtle stops pretending Recall Wilensky's four ways to see a circle. Euclid's: the set of points equidistant from a center — a *global* definition, requiring access to a point that is not on the circle. Descartes': x² + y² = K, an algebraic restructuration. The turtle's: move with constant linear and angular velocity and you trace a circle — and, crucially, **an ant walking the path can know it is a circle from inside, with a vanishingly small field of view, never once referring to a center.** And the emergent: many turtles, each given a random heading, each stepping forward by the same amount, *become* a circle no one drew. Wilensky offers the turtle's locality as a pedagogical kindness — it makes Newton's hard-won move from local properties to global form graspable by a child. We want to take it more seriously than that. **Locality is not merely easier to learn; it is the defining property of a decentralized system.** A world whose agents act only on what is in front of them, with no global coordinator and no privileged observer who can see the whole, is precisely the architecture of a distributed, fault-tolerant, sovereign system. The turtle was always a blueprint for how to build without a center. We simply kept the blueprint in a sandbox and called it a model. Release that constraint and the four circles become four eras of computing: the centered definition is the mainframe; the algebraic equation is the closed formula in the data center; the local turtle is the edge node that knows its neighborhood; the emergent circle is the commons that arises from their interaction. The restructuration is the same one Wilensky described. It has only changed venue, from the screen to the world.
## 3. Beyond simulation: computation as the new ontology Here is the line we wish to draw most sharply, because it is where we go past Wilensky rather than merely after him. Agent-based modeling *simulates*. A GasLab molecule is a stand-in for a real molecule; the model is a map, valuable precisely because it is not the territory. This is the right and honest stance for a science of understanding. But the agents we now deploy are not maps. A bead-agent at a URI is not a simulation of an actor — it *is* an actor: it has a body, a lifecycle, an address other agents can reach, and consequences that are real and not rehearsed. The map has become the territory. The turtle has stopped pretending. When that happens, computation stops being a description of a system and becomes the system's very mode of existence — an ontology, not a methodology. The structure *is* the computation: the current shape of the running namespace is not a record of an answer stored in a database, it is the answer, the way a soap film *is* the solution to its boundary problem rather than a depiction of one. This is least action made executable. Andreas Hübler's chains of ball bearings, polarizing in a viscous field and growing along the path of least resistance to close a circuit, find the same least-action topology that rivers, slime molds, and dendrites find — different physics, one geometry. A field of deployed agents minimizing their own action is a computational instance of that same law. The water-rights commons of the acequia is its social instance; Hübler's wires are its electromechanical instance; the agent ecology is its digital one. They are the same problem. This is also where a long collaboration with Stuart Kauffman informs the thread rather than decorating it. Kauffman's autocatalytic sets and his adjacent possible describe a universe that *constructs itself* — that brings actuals out of potentia through local, self-reinforcing interaction, with no blueprint and no builder. Agent-oriented architecture is that stance taken as an engineering commitment: you do not specify the global system and then implement it; you constitute local agents and let the structure they need come into being. Computation-as-ontology is the universe's own trick, ported.
## 4. `myself`: the agent that knows who asked There is a small piece of NetLogo and AgentScript syntax that turns out to carry the whole argument. Inside `ask agentset [ ... ]`, the word `self` refers to the agent currently executing — the turtle being asked. The word `myself` refers to the agent that *did the asking* — the caller, reaching in. In a simulation, `myself` is a convenience: a way to read the asker's variables during an iteration. In a deployed architecture it is far more. `myself` is **identity, provenance, and authority** — the agent's reference back to whoever invoked it. When a scheduler asks a population of agents to take their step, each agent still owns its own agency (it steps *itself*; nothing reaches in and mutates it), and yet it holds, in `myself`, a live reference to the one who asked. That back-reference is the exact seam where agent-based *modeling* becomes agent-oriented *architecture*: `ask` is no longer a loop, it is a request travelling over a link, and `myself` is the *other end of that link*. Delegation now carries its lineage; a step can answer its caller; trust and capability flow along the same edge that the invocation did. It is worth saying plainly because it is easy to miss: the World asks, but the turtle moves itself, and it never forgets who asked. That is not a footnote about scoping. It is a theory of agency compact enough to fit in a keyword.
## 5. The lesson from industry: Owen Densmore and the deployed turtle None of this is speculative for us, because we have watched it be a life. Owen Densmore is an early flower child of the internet and of personal computing — formed in the years before the network was commercialized, when the machine was a tool for expanding the individual and collective mind rather than a channel for extracting attention. That ethos — John Perry Barlow's ethos, of which more below — is not nostalgia in Densmore's case; it is a working method. His path runs through the places where interactive computing was actually invented and shipped: Xerox in its formative years, the Apple Lisa, the original Macintosh design team, Sun's research labs, and then, for a quarter century, the Santa Fe Institute summer school and Redfish — the last decade of it carrying AgentScript. The throughline matters more than any single stop. Densmore is someone who *shipped real interactive systems* and *taught complex systems at SFI* — and that combination is exactly what the science of restructurations needs and, by Wilensky's own account, cannot find a home for. AgentScript is the proof of concept. It is not a toy re-implementation of NetLogo; it is the turtle made *deployable* — a Model-View-Controller architecture in the browser, now with a WebGPU compute substrate, edge-native, no installation, addressable on the open web. Wilensky gave the turtle to the learner. Densmore gave it to the deployer. This is the lesson from industry the academy keeps mislaying: **deployment is the crucible that turns a representation into an architecture.** A representation that only ever runs in a classroom is judged by whether it is learnable. A representation that must run in production — under real load, real failure, real adversaries, real users who did not read the manual — is judged by whether it is *true enough to build on*. The agent paradigm passes that harsher test. We know because we have been building on it.
## 6. Against the weary giants: collective intelligence, not collected intelligence In 1996, from Davos, Barlow told the governments of the industrial world that they were not welcome in cyberspace — that the network was *not a public construction project* but "an act of nature" that "grows itself through our collective actions," a world whose order arose from "ethics, enlightened self-interest, and the commonweal" rather than from imposition, whose identities were "distributed across many of your jurisdictions," and whose only shared law was the Golden Rule. He was describing, thirty years early, the architecture this paper argues for: grown not built, local not centered, distributed in identity, collective in governance — a civilization of the Mind. The weary giants have changed costume. They are no longer only governments; they are the platforms. And the gravity of contemporary AI runs hard against Barlow's vision: intelligence is being *collected* — pulled into a handful of data centers, distilled into a handful of models, rented back to us through a metered pipe. It is a centered architecture in the precise sense of §2: a single privileged observer who can see the whole, and means of enforcement we have real reason to fear. The thing conspicuously *missing* from the Anthropic, Gemini, and OpenAI space is the very thing the turtle was always about — **linking between the nodes.** There are powerful minds in those data centers and almost no edges between them and us; there is collection without collective. We declare for the other architecture. Against intelligence collected, intelligence *collective*: many agents, massively linked, each sovereign on its own node, each acting locally, the global capability emerging from their interaction the way the circle emerges from the turtles. This is not a rejection of large models — bring them, as one rung among many — but a rejection of the *topology* that makes them landlords. The commons is grown through collective action or it is not a commons at all.
## 7. The call So: not a repository. A repository is a place you go to download a thing. We are not asking you to download a thing. We are asking for a restructuration to be allowed out of the room it was born in — for agent-oriented design to be taken as seriously in architecture as agent-based modeling has finally been taken in science; for the turtle to be deployed, not just demonstrated; for computation to be understood as a way the world can *be*, not only a way it can be *simulated*; and for the next generation of intelligence to be built as a civilization of linked, sovereign minds rather than a single mind that rents its attention. Wilensky and Papert opened a door and, with characteristic modesty, said the best way to serve the idea was to go deep on the one they had opened rather than spread thin across many. We honor that, and we walk through the door. The depth they opened in the classroom has a far side, and the far side is the world. Let the turtles loose.
### References & sources - Wilensky, U. & Papert, S. (2010). *Restructurations: Reformulating Knowledge Disciplines through New Representational Forms.* Constructionism 2010, Paris. - Barlow, J. P. (1996). *A Declaration of the Independence of Cyberspace.* Davos. (EFF, John Perry Barlow Library — verbatim text archived in this bead as `barlow-declaration-verbatim.md`.) - Papert, S. (1980). *Mindstorms.* (the Logo turtle.) - diSessa, A. (2000). *Changing Minds.* (the tick model; restructuration of kinematics.) - Kauffman, S. — autocatalytic sets, the adjacent possible; Kauffman & Guerin, *Did the Universe Construct Itself?* (the computation-as-ontology / self-construction thread). - Hübler, A. — self-assembling wires / ball-bearing chains in a viscous field (least-action topology; physical ground of the agent ecology). - Densmore, O. — AgentScript (`github.com/backspaces/agentscript`); NetLogo (Wilensky, CCL/Northwestern, 1999) as its lineage.