What Kernaro Assist Can Automate in Sparx EA (And What It Can’t)

The short version: Kernaro Assist is an AI assistant built into the Sparx EA client. It takes the mechanical weight out of modeling — element creation, diagram scaffolds, governance checks, and natural-language queries against the repository — all without leaving the tool. What it will not do is make architectural decisions, carry the stakeholder conversation, or invent quality that isn’t already in the model. Released in May 2026, it is still early; treat its output as a strong first draft, not a finished artifact.

What Kernaro Assist actually does

Assist works through a natural-language panel inside the Sparx EA client. Architects type a request or a question; Assist interprets it and either acts on the repository or returns information. Five capabilities are doing real work today.

Element creation from natural language. Type “Create a Capability element called Customer Onboarding under the Customer Engagement domain package, with MaturityLevel tagged as 2 and InvestmentCategory tagged as Transform,” and Assist creates the element, places it in the right repository package, and populates the tagged values. For bulk creation, this is markedly faster than clicking through the UI.

Diagram scaffolding. Give it a scenario — “Create an ArchiMate diagram showing the Customer Onboarding capability, the three applications that realize it, and the data entities they consume” — and Assist generates a first-draft structure. The architect then reviews, adjusts relationships, and corrects any misreads. The win is reaching a working starting point in minutes instead of building from a blank canvas.

Governance conformance checking. Assist can check a package or a set of elements against the MDG profile, flagging elements that are missing mandatory tagged values, ones built on base metaclasses where a stereotype should apply, or names that break convention. Done by hand across large packages, that is slow work; Assist compresses it.

Natural-language repository queries. An architect can ask “Which applications in the portfolio have a lifecycle status of Under Review and at least two capability realization relationships?” without writing a query or building a matrix. Assist traverses the repository and returns structured results — with accuracy that depends entirely on how consistently the relevant tagged values and connectors are populated.

MDG-guided tagging. When new elements are created, Assist suggests tagged-value completions drawn from the MDG profile and from how similar elements are already filled in. That chips away at the blank-field problem that drives inconsistent population in the first place.

The MDG dependency, stated plainly

This point is worth being blunt about. Assist reads the repository to generate everything it produces. Its suggestions about element types come from the stereotypes defined in your MDG profile. Its answers to repository queries come from the tagged values and connectors that are actually populated. Nothing it returns is better than what it reads.

If your MDG profile is undefined or applied inconsistently, Assist cannot infer what you mean by “application.” It sees a mixture of ArchiMate Application Components, UML Class elements pressed into service as application placeholders, and free-text notes. It cannot produce a coherent “list of all applications” because the repository has no coherent concept of one.

If your tagged values are sparsely populated, an answer to “which applications are at end of life?” returns only what happens to be filled in — which might be a fifth of the real answer, and misleading precisely because it looks complete.

None of this is a knock on Assist. It is how data-quality constraints propagate through any system that reads and interprets data. The same constraint governs EA GraphLink, the Power BI dashboards built on EA data, and every other downstream consumer of repository content. We make the same point at length in why your EA repository quality determines your AI output quality.

The practical implication: if your governance is weak, the highest-value use of Assist is governance improvement itself — using it to accelerate the conformance checking, tagging backfill, and element correction that lift repository quality. Not generating new content on a broken foundation.

What Kernaro Assist does not do

It does not make architectural decisions. Whether to retire an application, whether a capability is strategically important, whether a given integration pattern fits — these need judgment Assist does not supply. It can surface the information behind a decision (what depends on this application? what is its maturity rating?), but the decision stays with the architect.

It does not replace stakeholder engagement. Architecture is, at bottom, a communication discipline. Understanding what business stakeholders need, translating that into architecture, and presenting findings people can act on — none of that is automated. Assist changes the time available for that work by cutting mechanical overhead; it does not do the work.

It does not fix poor governance retrospectively. Assist can spot governance problems and speed up the cleanup, but it cannot answer a question that needs business context. Which of these three identically named applications is the canonical one? What lifecycle status belongs on a system whose original owner has left the organization? Those answers come from human inquiry.

It is still early. Assist was released in May 2026. Some interactions produce unexpected output, some requests get misread, and some features are still maturing. Review everything before it is used. The reliable team pattern is simple: Assist produces first drafts; architects review and finalize.

What Assist frees up time for

The real benefit is not automated architecture — it is the removal of mechanical overhead that currently eats architect time, freeing those hours for higher-judgment work. In most EA teams, a meaningful slice of the day goes to:

• Creating elements by hand and populating tagged values one field at a time
• Building routine diagram scaffolds from a blank canvas
• Checking large packages for governance conformance
• Compiling answers to ad hoc data questions — which applications connect to this service, which capabilities a given division owns

When Assist absorbs those, architects get more time for the work that genuinely needs their expertise: reading what the business needs, making trade-off decisions, engaging senior stakeholders, and designing patterns that hold up over time. We unpack where the hours actually go in what 70–80% of architect time actually goes on.

The multiplier framing is honest, but it cuts both ways. If an architect spends 40% of their time on mechanical tasks and Assist drops that to 15%, the time freed is real — and what the team does with it decides whether Assist delivers value. Time saved poured back into more mechanical modeling is a marginal gain. Time saved redirected into strategic architecture work is a genuine shift. That distinction is the whole argument behind AI Augmented Architecture.

Deploying Kernaro Assist effectively

Three things should be in place before you switch it on.

Frequently asked questions