The Bauhaus lasted fourteen years — 1919 to 1933. In that time it invented the discipline of modern design education. Not by teaching style, but by teaching materials. The Vorkurs (preliminary course) under Johannes Itten, and later Moholy-Nagy and Albers, required students to work directly with wood, metal, glass, clay, and textiles before they were allowed to design anything. The premise was simple: you cannot design with a material you do not understand. Understanding comes through friction, not theory.
The material is changing. The questions the Bauhaus was asking have not changed at all.
The material encounter.
Josef Albers had students spend a semester working with paper — just paper. Folding, cutting, scoring, layering. No glue, no tape, no external support. The constraint forced students to discover what paper could do structurally before they tried to make it do anything aesthetically. The aesthetic emerged from the material’s capabilities. Not imposed on them.
The paper exercise seems trivial until you try it. The material resists. Paper tears along the grain but not against it. It holds compression poorly but tension well. It buckles under load but a corrugated fold bears weight. The student who tries to impose a form on paper without understanding grain, weight, and structural limits produces something that looks designed but falls apart. The student who spends three weeks discovering what paper does produces something that looks inevitable. The difference is representational capacity — an internal model of the material that no textbook can transfer.
Yann LeCun’s JEPA architecture argues the same thing at the machine level: build the world model first. Learn the structure of the domain before generating output. The internal model — the representation — comes before the generation. Generation without representation is autocomplete. Representation without generation is understanding.
The parallel is exact. Albers wanted students to build an internal model of paper before designing with paper. LeCun wants machines to build a world model before generating predictions. Both are arguing that representational capacity is the scarce resource, and that generation is downstream of it.
In design education, this principle survived the transition from physical to digital. The Bauhaus paper exercise became the Photoshop exercise became the Figma exercise. The medium changed. The pedagogical structure — encounter the material, build the model, then design — remained.
Now the medium is changing again. And the pedagogical structure needs to change with it, or it stops working.
What Black Mountain added.
When the Nazis closed the Bauhaus in 1933, Albers went to Black Mountain College in North Carolina. The pedagogical experiment continued under radically different conditions — no institutional backing, no defined curriculum, no separation between the act of learning and the act of living. Students and faculty ate together, built the campus together, governed the college together. The material encounter extended beyond the workshop to the entire educational environment.
What Black Mountain added to the Bauhaus model was the insistence that the experiment and the education were the same thing. John Cage composed 4‘33” there — a piece where the performer sits in silence and the audience’s ambient noise becomes the music. Rauschenberg made his White Paintings there — canvases that reflect the room’s light and shadow, turning the environment into the artwork. Buckminster Fuller built his first geodesic dome there. These were not assignments. They were experiments conducted by people who had internalized the material encounter deeply enough to push into territory where the material itself was being redefined.
The agent-era design school might need both Bauhaus rigor and Black Mountain radicalism. The Bauhaus provides the pedagogical architecture: systematic encounter with materials in a structured curriculum. Black Mountain provides the experimental posture: the willingness to redefine what the materials are, to work at the boundary between known and unknown, to treat the school itself as a prototype rather than an institution.
The new materials.
If the Bauhaus workshops were organized around physical materials — wood, metal, weaving, pottery, typography — the agent-era design school would be organized around agent-era materials:
Attestation design. The material is the signed, verifiable claim. Students would need to work directly with attestation infrastructure — signing credentials, building claim graphs, revoking attestations, understanding what “verifiable” means at the cryptographic level. Not as theory. As hands-on material encounter. Fold the paper. Score it. Discover what it can bear.
What does it feel like to fail with attestation infrastructure? A credential chain that breaks because a root CA expired. An attestation that looks valid but was signed by a revoked key. A claim graph where a single compromised node invalidates an entire trust path. These failures are the grain of the paper — the material properties that only reveal themselves through direct encounter. Reading about key management is theory. Building a credential chain that survives adversarial testing is material understanding.
Capability manifests. The material is the structured declaration of what an entity can do. Students would build manifests for real organizations, test them against real agents, and discover what gets lost in the translation from human-readable description to machine-parseable specification. The gap between what an organization thinks it does and what it can declare in structured terms is the design problem.
Structured argument. The material is the claim graph — assertions linked to evidence, each signed, each with an explicit confidence level. Students would construct arguments as structured objects, not as prose. The discipline is different: prose allows ambiguity as a feature; structured argument treats ambiguity as a defect. Learning to think in claims rather than narratives is a material encounter with a medium that resists the habits of natural language.
Identity systems. The material is verifiable identity — DIDs, EUDI wallets, credential chains. Students would build identity systems for fictional entities, test them against adversarial scenarios (impersonation, delegation, revocation), and discover the design trade-offs between privacy and verifiability, between usability and security.
Machine-readable voice. The material is serialized brand voice — structured tone documents, constraint systems, few-shot example banks. Students would learn to express brand voice as structured data that a model can apply, and discover what survives serialization and what does not. The exercise reveals which aspects of voice are explicit and which are tacit, which can be encoded and which require human judgment.
Each material has its own grain, its own failure modes, its own moments of surprise. Attestation infrastructure fails silently — a revoked credential looks identical to a valid one until you check the revocation endpoint. Capability manifests fail loudly — an agent that cannot parse your manifest skips you entirely. Structured argument fails politically — making claims explicit invites challenge. Identity systems fail socially — privacy-preserving designs that are technically correct can feel hostile to users. Machine-readable voice fails aesthetically — the serialized version of a voice often sounds like a caricature of the original. Each failure mode is a material property. Each material property is a design constraint. The constraints are where the interesting work happens.
What the curriculum looks like.
A four-year agent-era design program might look like this:
Year one: Vorkurs. Material encounter with all five agent-era materials. No design projects. Just material exploration. Build attestations. Write manifests. Construct claim graphs. Create identity systems. Serialize voice. Fail, learn the constraints, fail again. The Albers paper exercise, but for digital trust infrastructure.
The Vorkurs would need to be genuinely uncomfortable. Albers’s original course was deliberately frustrating — students expected to learn to draw, and instead spent weeks folding paper. The discomfort was pedagogically productive: it broke assumptions about what design education was supposed to feel like. An agent-era Vorkurs that starts with JSON-LD, key management, and zero-knowledge proof systems would produce the same productive discomfort in students who enrolled expecting to learn visual design. The discomfort is the point. It means the assumptions are being challenged.
A concrete example: the first assignment might be to build a capability manifest for the student’s own university department. What does the department actually do? What claims can it make with evidence? What credentials back those claims? The student discovers that the department’s website says “world-class research” but cannot point to a structured, verifiable claim that an agent could evaluate. The gap between the marketing narrative and the attestable reality is the first material encounter. It is the equivalent of discovering that paper tears along the grain. Nobody told you. You found out by doing it.
Year two: Workshops. Specialized deep dives into each material. Students choose a primary and secondary material. Attestation design workshop. Manifest workshop. Structured argument workshop. Identity workshop. Machine-readable voice workshop. Each workshop has a practicing professional — not a professor — as the lead. The Bauhaus model: masters of craft, not masters of theory.
The workshop leads would need to come from fields that do not currently consider themselves design fields: cryptography, compliance engineering, knowledge representation, protocol design. The Bauhaus brought in practicing ceramicists, weavers, and typographers. The agent-era school brings in practicing cryptographers, identity architects, and regulatory technologists. The pedagogical role is the same: the master teaches the material, not the theory about the material.
Year three: Integration. Cross-material projects. Design a verifiable product passport (attestation + identity + structured argument). Design an agent-discoverable professional profile (manifest + identity + machine-readable voice). Design a trust surface for a marketplace (attestation + argument + identity). The projects are real — with real stakeholders, real agents consuming the output, real feedback loops.
The integration year is where the Bauhaus’s cross-workshop synthesis becomes critical. Gropius’s workshops were not silos — the weaving workshop used materials from the metal workshop, the typography workshop served the architecture workshop. The synthesis produced objects that could not have emerged from any single workshop alone. The agent-era equivalent is a trust surface that could not be designed by someone who only understands attestation, or only understands identity, or only understands structured argument. The integration is the design.
Year four: Thesis. An original design contribution to one of the agent-era materials. Not a research paper. A working artifact — a new attestation pattern, a new manifest structure, a new trust surface design. Evaluated by practitioners, not academics. Published as open-source. The thesis is the experiment. The experiment is the contribution.
What existing education gets wrong.
Current design education still assumes the output is pixels. The tools are Figma, the deliverables are mockups, the critique methods involve looking at screens and discussing visual hierarchy. This produces graduates who can design beautiful interfaces for human readers and have no vocabulary for designing structured data surfaces for machine readers.
The gap is not about adding a course on “AI and design.” It is structural. The entire pedagogical architecture — what counts as a material, what counts as a deliverable, how work is evaluated — assumes a biological reader. An agent-era design program would need to change all three simultaneously, which is why incremental curriculum reform is unlikely to work. You cannot graft the new materials onto the old architecture any more than you could graft machine weaving onto a fine arts curriculum and call it the Bauhaus.
Computer science education has a complementary blind spot. It teaches the cryptographic primitives, the protocol design patterns, the data structures that underlie trust infrastructure — but it does not teach how to design with them. The attestation is a data structure in a CS curriculum. It is a design material in the curriculum being described here. The difference is not semantic. It is the difference between knowing what paper is made of and knowing how to fold it.
The school as experiment.
The Bauhaus was not just a school. It was a proof of concept for a philosophy — that the separation of craft from art was destroying both — and every workshop, every assignment was generating evidence for or against the thesis. The school was the experiment. The graduates were the data.
The same structure might apply here. The thesis is that designers need to build representational capacity for agent-era materials before they can design with them, that this capacity can be taught through direct material encounter, and that people who undergo the process produce qualitatively different work from those who learn the theory without the encounter. An agent-era design school would be the experiment. The students’ work — evaluated by practitioners, tested against real agents, measured by real outcomes — would be the evidence.
The falsification conditions are concrete. If students who complete the material encounter curriculum produce agent-first surfaces that are indistinguishable from surfaces produced by engineers with no design training, the thesis is wrong. If the curriculum produces designers who can build attestation systems but cannot design trust surfaces that humans and machines both find usable, the curriculum is incomplete. If the workshop leads from cryptography and protocol design cannot teach material encounter to design students, the pedagogical model does not transfer. Each of these can be tested. The school would test them.
Whether the experiment is worth running depends on whether the machine reader is actually displacing the human reader as the primary interface consumer. The evidence suggests it is. If that transition is real, a design discipline organized around the new reader is not optional. It is the same kind of structural necessity that the Bauhaus recognized in 1919: the machine has arrived, the old education does not prepare for it, and someone needs to build the new one.
What survives from the Bauhaus.
The Bauhaus’s lasting contribution was not a style. It was a pedagogical architecture: encounter the material before designing with it. Build representational capacity before generating output. Ground the abstract in the physical. Let the material’s constraints shape the design.
That architecture survives the transition to agent-era materials. The materials are different — attestations instead of paper, manifests instead of metal, claim graphs instead of typography. But the pedagogical principle is the same: you cannot design with a material you do not understand, and understanding comes from working with it directly, not from reading about it.
Whether anyone builds this school is a different question. Design education is conservative, slow to adapt, and structurally resistant to the kind of fundamental curriculum change this requires. The Bauhaus itself was shut down by political forces hostile to its methods. Agent-era design education faces a different hostility: institutional inertia and the comfortable assumption that the existing curriculum — pixels, screens, user journeys — will remain relevant.
It might remain relevant. It will not remain sufficient. The machine reader is here. The materials are changing. The Vorkurs needs updating.
The Bauhaus opened three years after the end of one war and closed the year another began. Fourteen years to invent a discipline, train a generation, and disperse the methodology across the world through exile. The contribution was not the buildings or the chairs or the typography. It was the idea that design education starts with the material, not the brief. That idea survived the school’s destruction because it was true, and true ideas propagate even without institutions to host them.
The agent-era design school does not need to last fourteen years. It needs to articulate the idea clearly enough that it propagates: the machine reader is the new material, design education needs to teach encounter with it, and encounter means failing with attestations, manifests, claim graphs, identity systems, and serialized voice until the designer has built an internal model of what these materials can bear. The idea is simple. Building the institution around it is hard. But the idea might propagate without the institution, the way the Bauhaus idea propagated without the Bauhaus.
Whether anyone builds the school or not, the materials are already here. The designers who encounter them directly will produce different work from those who do not. The difference will be visible. That might be enough.