Manifesto for Next Information Architecture

The components IA must now address.

NEXT IA brings core information architecture grounding to in 4 key areas of intelligent agent design and development:

• Mental model alignment: Information architecture  create information experiences that incorporate human mental models not available to artificial intelligent agents. The gap between how humans understand information and how AI systems process is a structural IA not a UX problem to solve.
• Semantic integrity: Information architecture establish  meaning and relationships between concepts for AI systems to inherit and amplify. This is done with ontologies, taxonomies, knowledge graphs, and controlled vocabulary.
• Naming conventions: Human findability favors mental model alignment and language that matches user vocabulary. Machine findability favors formal semantic structures and entity disambiguation. They can conflict. A label that is intuitive to a human may be ambiguous to a retrieval system. Information architects can design for both simultaneously.
• Information governance: Information architecture governance provides a sole source of semantic truth across information systems with governance over: who decides what is authorized, what is findable, what is nameable? In an AI-mediated environment, these decisions become invisible as they are embedded in training data, retrieval weighting, and prompt design.

How We Got Here.

The information environment restructured itself.

Information and knowledge integrity of LLM-generated content remains subjective and requires semantic validation for coherence, completeness, and consistency. Information architecture decisions made in the design, development, and deployment stages propagate through to what the AI understands and synthesizes for users. Inadvertent programming bias can be amplified instead of filtered. This makes the accuracy and consistency of conceptual relationships more consequential than it has ever been.

In today’s intelligent information landscape, the information architecture is not always visible, and it is not always stable. Three structural shifts have fundamentally altered the territory IA must now navigate:

• The container is no longer the unit. LLMs retrieve and synthesize rather than navigate. There is no page for a user to land on. The information structure lives in embeddings, retrieval pipelines, and prompt design. The decisions about what exists, how it relates, and what surfaces are now happening at the infrastructure layer, not the view layer.
• Findability has bifurcated. There is human findability, navigation, search, wayfinding, and machine findability, structured data, semantic markup, entity disambiguation. These require different, sometimes competing, design decisions. Information architecture can serve both audiences for both infrastructure and view layers.
• The understanding of the goal of information outsourced to AI. Understanding is now mediated by a system with its own interpretation layer. A core information architecture principle is to produce user understanding by closing the gap between the information environment and the user’s mental model. When Richard Saul Wurman spoke of Information Architecture at 2018 IA Summit conference, he was speaking of ways of organizing systems and not necessarily how we interpret information architecture in Web and product design used today.¹ AI intermediaries now do the translation and that translation is not neutral.

The argument that AI makes IA obsolete inverts the existing dependency relationship. AI systems depend on knowledge structures to function reliably. Garbage-in, garbage-out applies at the conceptual level, not just the data level. Enterprise Information Architecture colleagues have a multitude of tools to build the semantic layer at its core delivering hallucination-free, AI-powered insights by leveraging an explicitly defined semantic model, ensuring transparency, traceability, and trust in decisions.

LLMs and the Content Management System (CMS) core.

The CMS premise of rows and columns is now structurally challenged. The classic CMS is a workaround for the limitations of relational databases. LLM-based systems can make traditional content types and classical taxonomies obsolete because AI systems can categorize content by phrase or find the right content based on a prompt.

Bob Boiko, author of The Content Management Bible,  starts with the actual problem — how do you manage and deliver content using available computational infrastructure — and follows it wherever it goes, including into territory that obsoletes his own prior work. He expands his understanding to meet the problem rather than expanding the problem’s definition to meet his understanding. This new architecture² relies on content chunks that are run through an embedding model that encodes content as vectors, stored in a vector database, retrieved by semantic query, merged with user information to create a prompt for an LLM, and the LLM responds with a consolidated content answer to the user prompt.

Information Architect Jason Hobbs created a framing of information architecture that he calls an Information Architecture Design (IAD). Hobb’s IAD rests on HCI’s (human computer interaction) “third wave” of embodiment, culture, and the wild.² However, IAD lacks technical depth, or consideration for tangible computing, ubiquitous systems, AI interaction research, and accessibility engineering. Blended spaces and embodied information experiences are not frameworks that AI architects, knowledge graph engineers, or RAG system designers can use to programmatically produce human-centered applications for the new intelligent environment.

In Andrea Resmini’s and Luca Rosati’s Pervasive IA³ the experience of ecosystems work is grounded in the phenomenology of how people experience information across blended physical-digital spaces. It describes the meaning of making the new environment without engaging with what is producing it: machine learning infrastructure, embedding models, recommendation systems, platform architectures, etc.

The result is a discipline definition that may be intellectually respectable within humanities and social science frameworks but has little relevance to those making structural decisions for applications that power the intelligent information landscape. It is information architecture as a discourse community rather than IA as a practice. Information Architecture Design, and Pervasive IA and the Polar Bear book are describing the view from the shore. The work to advance the practice to a more meaningful state is found deep in the water where the work is being done in real time.

What is in a name?

Names are the basis of all information systems. They are the containers that hold meaning and make it useful. They confine meaning with a boundary around bits of sense. In information systems, names make meaning into data. Information structures are systems of names. Information architects create information structures. Information architects are better suited to developing naming conventions that work across applications and disciplines.

LLMs do not eliminate the naming problem. They inherit it wholesale and amplify it. When an LLM retrieves and synthesizes, it operates on whatever naming system is embedded in the corpus. Good names propagate meaning. Names preserve what is genuinely essential about IA. Names let go of the implementation scaffolding (content types, taxonomies as bureaucratic artifacts, markup) that was always instrumental, not fundamental.

The Polar Bear book with its user-centered focus starts with the user mental model and finishes with design structures to match. NEXT IA starts with the information’s inherent structure and names. That is closer to how AI systems work — corpus-centered rather than user-centered. What neither the IA community nor the content management community has done is connect these two streams with the technical depth the current moment requires.

Bob Royce and System Thinking.

Bob Royce’s 2011 EuroIA presentation argument was a call to bridge exactly this gap from inside the IA field. He proposed that IA should fuse three historically separate streams.

• The information design tradition (Wurman and Xerox),
• The information systems tradition rooted in Shannon’s computational approach,
• Librarian traditions of Rosenfeld and Morville.

Fifteen years of not heeding the argument has left the field unable to engage at the layer that matters most. The result is technically functional systems with structurally bad information architecture with nobody in the room knows how to solve that problem.

Information Architecture as a System.

Complexity scientists, cognitive scientists, information theorists, political scientists, and defense analysts from UC Merced, the Santa Fe Institute, USC, Harvard’s Berkman Klein Center, Stanford, Princeton, the Naval Postgraduate School, and the Max Planck Institute see information architecture as a system in their work Information Architectures: Socio-Technical Frameworks and Global Dynamics (Smaldino, et al).⁴

Their definition is that information architecture is the set of structural features and affordances that enables and constrains information flow between sources and receivers within a population — including both physical infrastructure such as transportation networks, broadcast media, and telecom, and cultural features including social norms, laws, languages, and both formal and informal institutions, as well as digital infrastructures and algorithmic technologies.

The Smaldino et al. framework expands the traditional conception of IA across four critical dimensions:

• Foundation models as information architectures: Under this definition, a foundation model is not a tool used within information architecture. It is itself an information architecture. Its training data, its attention mechanisms, its Reinforcement Learning Human Feedback (RLHF) alignment — these are structural features that enable certain information flow and constrain others.
• Knowledge graphs and ontologies as information architectures: A knowledge graph is a set of structural features that governs what concepts exist, how they relate, and therefore what information flows are possible between them. Under Smaldino’s definition, designing an ontology is IA work, not a subset of data engineering, not a UX exercise, but the primary architectural act of shaping information flow.
• Systems that learn: An adaptive RAG (Retrieval Augmented Generation) pipeline, a recommendation engine, a fine-tuned model. These are information architectures that are simultaneously designed and evolving. The governance challenge is not just what did we built but what it is becoming. No classical IA framework has machinery for that question. Complexity theory does.
• Competition among information architectures: In enterprise terms, this is the competition between organizational knowledge architectures — between the knowledge graph a company has built and the foundation model it has licensed, between its internal taxonomy and the semantic structures its AI vendor assumes.

NEXT Information Architecture

The current state of the IA community.

The IA community has lost its thought leadership. The Polar Bear book authors have moved into UX publishing and event management, or into niche areas of AI tool demonstrations. Resmini and Hobbs are focused on developing academic concepts with tepid application to the existing or emerging landscapes. The Information Architecture Institute ceased operations in 2019. The field is in maintenance mode without advancing its position in the AI conversation where it most needs to be present.

The Polar Bear book was always a subset — a useful, timely, accessible subset, but a subset, nonetheless. The practitioners trained by that book learned the vocabulary without the infrastructure. They learned what things were called without learning how the systems that carried them worked. The Enterprise IA tradition it drew from was broader, technically deeper, and more structurally complete.

That tradition continued its development with knowledge management, information systems, Enterprise architecture, knowledge graphs and ontology engineering. NEXT IA reconnects the field to this fuller tradition that it was derived from and extends it to account for what AI systems have changed.

What the field must become.

The originating canon for information architecture depends on Web IA and content organization for human users of designed digital interfaces.

NEXT Information Architecture adds cross-discipline infrastructure by providing ongoing education on how intelligent systems work, complexity theory for understanding emergent behavior, and Soft Systems Methodology to create organization change that produces a shared vision from all stakeholders and disciplines to ensure human factors are considered at the beginning of the design phase, not retrofitted at the end. The NEXT IA goal is to build a cross-discipline community that includes people with the technical depth Web IA never developed, pairing it with the human factors conceptual frameworks — SSM, complexity theory, CATWOE and semantic integrity that the AI and data engineering community lacks.

Information Architecture is enterprise information architecture applied to the intelligent information landscape.

The two domains share conceptual foundations, both require semantic precision, both involve governance of naming and relationships, both are now directly implicated in how AI systems behave.Today, architectural decisions are being made in real time. The costs from the lack of NEXT information architecture are measurable. The window for establishing the field’s relevance is closing rather than opening.

The NEXT IA canon.

NEXT IA operationalizes the full-scope Smaldino et al. definition by providing the cross-discipline practice methodology, governance frameworks, human factors integration, and community infrastructure that neither Smaldino’ s theoretical frame nor the existing IA field’s practice tradition provides on its own. This is accomplished with:

• Complexity theory as theoretical grounding for systems that operate in a dynamic environment with constant learning and change.
• System thinking that embraces algorithms and machine mediation as a user classes.
• Soft Systems Methodology’s CATWOE (Customers, Actors, Transformation, Worldview, Owners, Environment) for organizational change tools
• The ECMGME framework (Educate, Collaborate, Model, Generate, Measure, Evolve)
• Enterprise IA governance frameworks applied to intelligent information systems.
• Cross-discipline community infrastructure connecting IA with computer science, engineering, content strategy, UX, and SEO

The Smaldino et al. definition names what information architecture can be for the intelligent landscape by bringing structural governance of information flow, deliberate and emergent, at organizational and societal scale, with AI systems as first-class participants. This becomes the full scope that Web IA has been avoiding for many years and is now made necessary by the systems that have taken over information delivery and consumption.

What is next for NEXT IA.

Incorporate Enterprise IA as a model

Enterprise Information Architecture’s key domains are classification, labeling, search design, content structuring, and information governance. These work together to create a coherent information environment that enables digital platforms, business functions, and workflows to operate with aligned and reliable information. Absent from that framing are end users, navigation, sitemaps. Machines now mediate wayfinding both inside and outside the enterprise.

Sixty-one percent (61%) of companies admit their data assets are not ready for generative AI due to unstructured, trapped in silos, or poor-quality data. Seventy percent (70%) find it hard to scale AI projects relying on proprietary data.⁵  These are not a data engineering problems. They are information architecture problems. The data exists. It is ungoverned, inconsistently labeled, and semantically incoherent. AI systems can retrieve it, but they cannot interpret it reliably because the conceptual structures that would give it meaning have not been established and maintained.

Data engineers, machine learning engineers, and enterprise architects are making decisions that are fundamentally information architecture decisions: What gets classified. How concepts relate. What is canonical. All are currently accomplished without information architects in the loop. The engineers do not understand that they are solving retrieval problems without solving underlying information architecture problems.

The deeper governance problem is structural: what concepts are canonical, who owns them, how conflicts between business units are resolved, how the information environment evolves over time. NEXT IA will be a framework with privacy, fairness, accountability, and sustainability as non-negotiable foundations.

Build the cross-discipline community.

The goal for NEXT IA is to supply practitioners who combine IA and human information behavior foundations with a deeper understanding of AI infrastructure — practitioners who know where the structural problems live and can contribute more human-centered methods to address them.

This is not accomplished by updating reading lists or adding AI modules to existing IA curricula. It requires building a new cross-discipline community from the ground up — one that structurally includes the technical disciplines IA has been absent from and provides the methodological framework to make cross-discipline collaboration productive rather than performative.

The NEXT IA ECMGME framework.

The NEXT IA methodology is structured around six phases that move from cross-discipline education through collaborative design to adaptive governance. The framework is designed to be practiced by mixed teams of IA, content strategy, UX, SEO, computer science, and software engineering practitioners working together on real AI systems.

ECMGME is not a linear process. Findings in later phases regularly require cycling back to earlier ones. This is not failure — it is the methodology working as intended. Complex systems cannot be fully understood in one pass, and the information environments they operate in change. The Evolve phase ensures that the framework itself improves with every engagement, raising the rigor of the community’s practice over time.

Socializing the framework series.

The first act of NEXT IA is education to introduce the core theoretical components to practitioners across all six disciplines simultaneously. The NEXT webinar series runs over five sessions:

• Introduction to NEXT: The case for cross-discipline collaboration. Why the information landscape has changed, what it requires from IA, and what the NEXT community is building.
• Complexity Theory: How complex adaptive systems behave. Feedback loops, emergence, attractors, and sensitive initial conditions — the conceptual foundations for understanding AI content systems as systems that learn and evolve.
• Systems Thinking for NEXT: Causal loop diagrams, stocks and flows, leverage points. How to model the dynamic behavior of information systems before they are built.
• Organizational Change Management with SSM: Soft Systems Methodology and CATWOE for navigating the messy human-activity systems that AI design decisions are embedded in. How to surface worldviews, build shared vision, and manage the change that AI systems require.

A NEXT Gathering

The webinar series builds toward a virtual conference that brings all six disciplines into cross-discipline education and dialogue with presentations from individual discipline thought-leadership and cross-discipline panel discussions.

• Workshops: Full-day NEXT practitioner labs. Mixed-discipline teams apply the ECMGME framework to real or constructed problems in real time. This is where cross-discipline practice begins rather than being described.
• Main program: Opening keynote (cross-discipline bridge talk), individual contributor sessions from IA, content strategy, UX, and SEO thought leaders, and a closing panel with Q&A focused on specific collaboration opportunities and working group formation.

The closing panel on Day 2 includes a structured exchange in which computer science and engineering practitioners identify what human factors practitioners need to know to participate in AI development from the earliest stages — conceptual foundations, shared vocabulary, and specific learning pathways. This exchange is reciprocal: human factors practitioners articulate what technical practitioners need to understand about how information systems behave as human activity systems.