How to Make Architecture Visible to Non-Architects
EA teams spend enormous effort building models that stakeholders can't access or use. Here's how AI is changing…
Read →A framework for sustainable technology adoption in mid-size organizations.
A practitioner’s guide to the six stages of EA (Enterprise Architecture) evolution and the shift that changes everything. By Ryan Schmierer, Sr. Managing Partner, NovoCircle · March 2026
Enterprise architecture did not begin with TOGAF (The Open Group Architecture Framework). It began the first time a human organization became too complex for any single person to hold in their head, and someone had to build an external representation of it so the organization could understand and govern itself.
This paper argues that enterprise architecture has evolved through six recognizable stages, each one lifting a constraint the previous generation accepted as permanent. Those six stages are: The Carried Map, The Age of Diagrams, The Connected Repository, A Common Language, The Governed Model, and Architecture Without Amnesia. Each transition followed the same pattern — a capability arrived that made the previous limitation look like a choice rather than a necessity.
The sixth stage, Architecture Without Amnesia, is the most significant shift in the arc. Every previous stage improved how well architecture was captured and communicated. This stage changes what architecture can know. Before this shift, every repository in the world suffered from the same structural limitation: it preserved conclusions, not reasoning. AI (Artificial Intelligence) ends that era. The repository can now retain the source material, the reasoning, and the organizational context alongside the model elements they produced.
| Stage | Name | Organizing Idea | What Triggers the Transition |
|---|---|---|---|
| 1 | The Carried Map | Architecture exists only inside one person | A retirement, resignation, or failure exposes the invisible map |
| 2 | The Age of Diagrams | Architecture survives the meeting, but not the year | Diagrams are always out of date and cannot be trusted for decisions |
| 3 | The Connected Repository | Elements have identity; the model has memory | The repository is growing but becoming harder to trust as it grows |
| 4 | A Common Language | Architecture becomes portable across people and organizations | Language is consistent; structural choices are not |
| 5 | The Governed Model | The repository earns the right to be trusted as data | Opportunity signal: repository is clean, AI tools are ready |
| 6 | Architecture Without Amnesia | AI ends the era of conclusions without reasoning | Capacity constraint is visible; the transition is technically possible now |
There is an architect in almost every organization at this stage, even if nobody calls them that. They carry an extraordinarily detailed map of the organization’s technology landscape. They know which systems interact, which integrations are fragile, and which decommissioned component is still silently receiving traffic. The map is real. It is accurate. It is also invisible — it exists nowhere except inside that person.
The first change required to leave Stage 1 is cultural, not technical. The organization has to accept that architecture is a shared function, not an individual attribute. The commitment to maintaining something is more important than what that something is built in.
The most common Stage 2 tool is not Visio. It is PowerPoint. When an organization commits to capturing its architecture for the first time, it reaches for the tools it already knows how to use. These tools share a fundamental limitation: a box is a shape. It has no identity, no type, no properties, and no persistent relationships. Change one instance and every other instance stays wrong. The result is the pathology every Stage 2 practitioner recognizes: diagrams age into fiction.
The move from Stage 2 to Stage 3 is a conceptual shift as much as a tool shift. In Stage 2, a box is a shape. In Stage 3, a box is an element — an object with a unique identity, a defined type, properties that describe it, and relationships that connect it to other elements. That element exists once in the repository and can appear on any number of diagrams. The diagram is no longer the artifact. It is a view of the underlying model.
The EA profession settled on shared modeling languages — ArchiMate, UML (Unified Modeling Language), BPMN (Business Process Model and Notation), SysML — that now define the standard vocabulary of the practice. The shift from Stage 3 to Stage 4 is a shift from private vocabulary to shared vocabulary. A new architect joining from a different organization can review existing models and contribute new ones within days: not because they have credentials, but because the language exists independently of any single team.
Stage 5 is the most complex to describe because it encompasses two meaningfully different internal positions. An organization at the entry of this stage has adopted reference frameworks, established governance processes, and built architecture practices that look mature from the outside. An organization at the far end of this stage has something more specific: a repository whose contents are semantically consistent, machine-readable, and trustworthy as a data source. These are not the same thing.
The gap between them is invisible until you try to do something automated with the repository — and then it becomes the only thing that matters.
Every EA repository ever built has suffered from the same fundamental limitation. It is a lossy compression of organizational reality. When an architect builds a model, they synthesize an enormous volume of input — conversations with stakeholders, observations of running systems, project documentation, trade-off analyses, historical decisions — into the model. The synthesis is the architect’s central professional act. It is also an act of destruction. The reasoning is discarded. Only the conclusion survives.
AI removes this constraint. AI systems can ingest, retain, and continuously process source material that has always been available but never incorporated into the repository. The result is a repository at Stage 6 that is not a snapshot of what the architecture team believes the organization looks like. It is a living representation of what the organization actually is: how its technology landscape is structured, why it is structured that way, and what the architectural implications of current changes are. This is architecture with memory. Architecture without amnesia.
The resistance is the signal. Every stage transition in this arc was resisted by people who were successful in the current stage. The diagramming practitioner who resists the modeling tool is not being irrational. The resistance is evidence that the current stage is delivering real value, which is exactly when the constraint of that stage is hardest to see.
The stage cannot be skipped but the pace can be accelerated. The organization that jumps from Stage 2 to Stage 5 because a consultant configured a sophisticated tool will find that Stages 3 and 4 still need to be done. But the pace of transition can be accelerated. AI is now accelerating the later transitions in a way that was not previously available.
Governance is only as strong as the model. Every major maturity model measures organizational governance processes. None of them measure whether the governance is encoded in the model itself. A governance constraint encoded in the model’s validation rules cannot be bypassed without someone actively choosing to bypass it. For AI augmentation, only this second kind of governance works.
The next transition is already possible. The capability that enables Stage 6 exists now. AI systems can ingest organizational context at scale. Repository connections to AI environments are technically straightforward for a practice that has completed Stage 5. The question is not whether this transition will happen. It will happen across the profession. The question is whether it happens by design or by drift.
The Long Arc is NovoCircle’s enterprise architecture maturity framework. It describes six stages of EA evolution — from Stage 1 (The Carried Map) to Stage 6 (Architecture Without Amnesia) — and explains what each stage delivers, what eventually makes it insufficient, and what the next transition actually requires.
Most mid-size organization EA practices sit between Stage 2 and Stage 5. The diagnostic question is not “what tools do we use?” but “what has started to feel like an unavoidable limitation that we’ve just learned to live with?” That answer almost always points to the current stage’s ceiling. Read our self-assessment guide to locate your practice on the arc.
Every NovoCircle EA engagement begins by locating your practice on the maturity arc. We identify the constraint on the next transition, assess what the Stage 5 to Stage 6 preconditions look like in your specific environment, and return a prioritized set of recommendations with effort estimates and sequencing guidance.
The framework is most usefully applied to a specific EA practice and its immediate organizational context, not to the enterprise as a whole. A Stage 5 corporate EA function can coexist with Stage 2 PowerPoint diagrams in business unit SharePoints. This is normal and does not invalidate the assessment.
Every NovoCircle engagement begins with a fixed-scope assessment — locating where your practice sits, identifying the constraint on the next transition, and returning a prioritized set of recommendations. Fixed scope. Fixed price. No agenda beyond getting it right.
The central conclusions from this research, presented for practitioners and technology leaders.
The majority of automation projects that do not deliver expected ROI failed not in implementation but in the scoping and assessment phase — organizations built before they understood.
Organizations that conducted independent platform assessments before procurement achieved meaningfully higher implementation success rates than those led through vendor evaluation processes.
Over half of surveyed deployments reported meaningful underutilization at twelve months post-deployment. In nearly all cases, structured training had not been scoped as part of the project.
Organizations that prioritized foundational data and integration work before building automation or analytics layers reported significantly better sustained outcomes.
The complete paper includes methodology notes, detailed findings, practitioner recommendations, and a framework for applying these insights to your own technology investment decisions. No registration required.
Download PDF (Free)PDF · Free · No email required