The Ontology Pipeline™, Refresh

Where We Were, Where We Are and Where We're Headed

When I published the original Ontology Pipeline™ framework in January 2025, I wanted to do one thing—make the invisible visible. The work of building semantic knowledge management systems had been treated as a black box for too long — expensive, undefined, and nearly impossible to scope or defend to a skeptical stakeholder. The pipeline gave it shape, vocabulary, sequence and logic.

That was the beginning.

What has happened since is something I did not fully anticipate, even if I sensed it coming. The demand for semantic knowledge infrastructure has gone from a quiet argument made in the back rooms of knowledge management and indie conferences to a front-page business imperative. And with that demand has come a flood — of confusion, shortcuts, AI-generated taxonomies presented as strategy and vendor marketing dressed up in ontology language. The work has not gotten easier. If anything, it has gotten harder to do well, and easier to do badly.

This is a refresh. Not a revision of the framework — the pipeline still holds — but a reckoning with where we are now, and what it demands of us.

Where We Were

The original Ontology Pipeline™ was born from a frustration I had carried for years and codified after successful implementations across six institutions over a ten-year time span. I observed Organizations were failing at knowledge management and semantics not for a lack data, but because they lacked structure. They rushed to build taxonomies without first establishing controlled vocabularies. They reached for ontologies without the foundational data hygiene that makes ontological reasoning possible. They tried to write the story before they understood the characters or the scene.

The framework I proposed drew directly from library and information science and the Semantic Web— domains that has been organizing knowledge since long before the term artificial intelligence entered the popular lexicon. Librarians solved these problems and Semantic Web practitioners further developed the pragmatic solutions . They built principled systems for disambiguation, semantics, hierarchies, ontologies and metadata standards. They codified what it means to describe a thing, to classify it, and to make it findable by both humans and machines. The Ontology Pipeline™ translated those principles into an enterprise-ready, repeatable framework.

The pipeline’s stages — controlled vocabulary, metadata standards, taxonomy, thesaurus, ontology, knowledge graph — were intentionally designed to follow a logical workflow, where one stage in the pipeline prepares the ground for the next. You cannot build a sound taxonomy on the heels of an undefined vocabulary. You cannot build a thesaurus without a taxonomy from which to mature. You cannot build an ontology on data that lacks logical integrity. And you cannot build a knowledge graph that reasons reliably without all of the above in place.

The framework existed because without it, organizations had no way to scope the work, estimate investment, or prove value. Knowledge management operated as an afterthought — a solo expert blamed when AI implementations failed, while the real problem sat upstream in the data infrastructure itself. I created the framework because I was often that solo practitioner, unable to move semantic knowledge infrastructures forward without presenting the work-to-be-done in manageable segments that translate into business value.

Where We Are

The energy around semantics, context, taxonomies and ontologies has reached a frenetic pace as the signal-to-noise ratio has dropped considerably.

The public release of ChatGPT in November 2022 did not create the need for semantic knowledge infrastructure— that need existed long before. What it did was make the absence of that infrastructure impossible to ignore. Organizations scrambled. Leaders who had never thought seriously about controlled vocabularies suddenly needed knowledge graphs. Teams that had spent years managing taxonomies in spreadsheets were asked to support RAG implementations overnight. And into that gap poured a wave of products, opinions, and overnight experts, all claiming to have the answer.

Some of what has emerged is earnest and honest. The field has produced serious work, and there are organizations moving the semantic needle. Ontology engineering has no doubt, matured as a discipline. Standards like SKOS, OWL, and RDF have found their way into more data infrastructures.

The conversation about symbolic AI — the structured, rule-based knowledge representation that library science and the Semantic Web communities have practiced for decades — has finally arrived in enterprises. Researchers and practitioners are rediscovering what information scientists have known all along—language models hallucinate when they lack grounded, structured knowledge, and that structured knowledge requires intentional human effort to build.

But the noise has become intense. Cookie cutter AI taxonomies are being sold as knowledge strategy. The term ontology is now co-opted by vendors, to describe product category trees or simply, words with meaning. Context graphs have emerged as a buzzword, attached quickly to job titles and startup flyouts, without methodology. And the accusations of gatekeeping have grown louder — the claim that expertise in knowledge organization is an artificial barrier—with attempts to dissolve the barriers to entry by implementing AI-generated taxonomies and ontologies, on demand.

I have addressed this confusion directly in my writing, and I will say it plainly here—a structurally invalid taxonomy is not a taxonomy. An AI-generated controlled vocabulary without definitions, without deduplication, without a methodology for resolving synonyms and homonyms, is not a controlled vocabulary. It is a list. Lists are not knowledge infrastructure. And the failures that result — AI systems that cannot find what users need, RAG implementations that hallucinate because they lack semantic grounding, enterprise search that returns invalid or irrelevant results — these failures are predictable consequences of skipping the work.

The work cannot be skipped. It can be accelerated, and AI can be a wonderful partner in acceleration. But the work itself — the thinking, the defining, the modeling, the validation — requires human judgment. It requires expertise and the discipline of asking competency questions before writing a single class declaration. It requires understanding that difficulty articulating a definition is not a reason to proceed with modeling — it is a signal to stop and think.

And For Some, Success

There are practitioners and organizations that have taken the Ontology Pipeline™ to heart and have done the work. I have met with these people, their organizations, and learned first hand, of their successes implementing the pipeline framework. And it works. The Ontology Pipeline™ breaks down the work into composable segments, much easier to sell to a business and fit into roadmaps with deliverable assets at the end of each phase of building.

The feedback and response from successes are met with ‘thank you’, as one of the biggest obstacles to organizational acceptance and adoption tends to be, “I do not know how to get started” or “I do not know how to sell this project to the business”.

So the Ontology Pipeline™ does work, and I have celebrated in the successes of multiple practitioners and enterprises. But there is still the skills problem—that has not gone away. People are hungry for learning and skills—yet organizations are not providing the opportunities for learning ahead of layoffs. And the fear of losing one’s job for lack of AI-ready skills is creating conditions rife with pressure and anxiety—an unfair predicament for many.

The Hard Work is the Elephant in the Room

We have an education problem.

The demand for semantic engineers, ontologists, and knowledge architects has never been higher. The supply of people who genuinely know how to do this work has not kept pace. And the gap is being filled by people who have encountered the vocabulary without the methodology — who know what a knowledge graph is supposed to look like but not how to build one that reasons correctly and survives an organization that does not hold still.

This is not a criticism of those people. It is a description of a basic failure in how the field has been taught — or rather, how it has not been taught. Library and information science and the Semantic Web have the principles. Enterprise data practices have the organizational context. But these domains have remained siloed, each speaking its own dialect, assuming a background and identity that is defined by pre-AI conditions.

The answer is not another certification that teaches vocabulary without practice. And it is not a tutorial that walks someone through a tool interface without explaining the reasoning behind the modeling decisions. It is investment in education, learning and mentorship. The kind of education that puts people in front of real problems with real data, guided by practitioners who have built these systems and experienced great success and watched them fail, both enriching experiences that make for great practitioners.

Leaders and organizations must invest in learning, in upskilling , as this simple gester is one of compassion. When so many people fear losing their livelihoods with threats of roles becoming obsolete, it is not enough to tell workers to upskill without providing the educational opportunities.

The field needs people who can do this work — not people who can describe it, but people who can sit down with a messy enterprise vocabulary, identify the synonyms and the homonyms and the orphaned concepts, build a metadata standard that normalizes entity-value pairs, mature a taxonomy into a thesaurus using SKOS, and then — only then — begin the ontology build. People who understand that governance is not documentation and that the day an ontology is deployed is the day it starts to drift. That maintenance is not a phase at the end of a project — it is the project.

Teaching this is hard. Learning it is harder. Breaking old habits — the instinct to rush to the knowledge graph before the vocabulary is clean, the temptation to treat AI output as finished work rather than raw material — requires discipline, humility and practice. When organizations and leaders stand behind learning, they support the learning process while investing in workers. People feel safe, supported and able to explore new ways of operating, building and growing. Curiosity flourishes and fear diminishes.

The knowledge tools exist and the standards are mature. The business cases have never been so clear. Will we build the knowledge infrastructure this moment calls for or will we keep papering over its absence, with products that look like knowledge but cannot logically reason?

Where We Are Headed

The Ontology Pipeline™ was designed as a framework for building. What I see ahead is a pipeline extended by two disciplines that have always been alluded to but quietly made explicit—governance and AI partnerships.

Governance is the engineering practice that keeps an ontology coherent across change. Organizations evolve and workflows shift. AI systems are integrated, upgraded and retired. New competency questions emerge that the original model cannot answer. Without a formal governance structure — versioning conventions, ownership protocols, change management processes, validation matrices — even a well-built ontology degrades. Governance is a living discipline, and it must be treated as one.

AI partnership is the more nuanced area to explore. AI is already participating in ontology work — in entity extraction, gap analysis, drafting candidate vocabularies for human review and populating ontologies for validation. This is productive work and a welcome partnership for knowledge infrastructures and the Ontology Pipeline™. Working with AI productively, using knowledge means not delegating modeling judgment to the machine, but using the machine to accelerate the human work — to surface candidates, to flag inconsistencies, to run reasoners, to test SPARQL queries against competency questions.

This is a critical point to clarify for the industry. AI that generates a taxonomy wholesale, without human review, unable to validate against standards, missing design heuristics such as competency questions and coverage models—that AI is producing a liability disguised as an asset. AI that assists a trained semantic engineer in building a better ontology, faster is just plain smart.

The revised Ontology Pipeline™ is a structured knowledge infrastructure , the same original five stages, governed by formal processes, accelerated by AI tools that augment rather than replace expert judgment, and staffed by people who have been trained to do the work. The same framework with the addition of governance and AI collaboration.

The questions driving forward, are the ones I have always asked:

What problems are you hoping to solve for?

What is this knowledge system designed to answer?

Who can change what, when and how?

How can we collaborate across disciplines?

How will you sustain this work with skills and people?

Those are organizational questions— and the hardest kind, because they require not only tools and standards, but accountability, investment, and the willingness to treat knowledge infrastructure as the backbone of AI systems and workflows.

The Ontology Pipeline™ framework has always been about making the work legible — to stakeholders, to engineers, to anyone trying to understand why building knowledge infrastructure correctly requires time, expertise, and an iterative, disciplined methodology. That remains the same.

What has changed are the stakes and the urgency. There are people hungry to learn how to build knowledge systems and the Ontology Pipeline™ is a framework designed to help people learn and understand how to build and implement controlled vocabularies, metadata schemas, taxonomies, thesauri, ontologies and knowledge graphs. What the work looks like, how to measure each knowledge asset, how to block out the work, in what sequence, to what end.

The Ontology Pipeline™ provides the framework to realize semantic knowledge infrastructures. We are adding governance and AI-augmented processes, an update to bring the framework into the now. To bring the framework into reality at scale, we need leader-supported upskilling—education—so we can build for reliable AI with new knowledge practitioners, for tomorrow.

---

More From Intentional Arrangement