Kernaro Assist or Claude with AI Power Tools: which fits which job

An architect moving from manual modeling to AI-augmented work wants the AI to build and correct the model while they decide what it should say. Two tools now let an AI read and write a live Sparx Enterprise Architect model: Kernaro Assist inside EA, and Claude working through AI Power Tools from outside it. They overlap on modeling and analysis, and each fits some jobs better than the other. Either way, the AI is only as good as the person directing it.

A disclosure first. Kernaro Assist is built by Sparx Systems, the vendor of Enterprise Architect, and Sparx Services is independent of Sparx Systems Pty Ltd. AI Power Tools for Sparx EA is published by NovoCircle, of which Sparx Services is a practice. We teach both. Both launched in 2026, so neither has a long track record, here or anywhere else. Judge them on what they do today.

Same repository, different shape

Assist lives where the modeler already is. There is no second window and no context switch, which is a real advantage when modeling is most of the day. Claude with AI Power Tools lives in the AI platform and reaches into EA through the server, so the same session that reads your repository can also read a spreadsheet, a wiki page or anything else your organization has approved as a connector.

The underlying model does not separate them. Kernaro is bring-your-own LLM. You supply the keys for Azure OpenAI, OpenAI, Anthropic Claude or an enterprise endpoint, and you pay the provider for usage. Claude with AI Power Tools uses Anthropic’s models through your organization’s Claude plan or API account. You can run Claude behind Kernaro Assist too.

Jobs where Kernaro Assist fits

• Day-to-day modeling by people who live in EA. AI help in the tool they already use saves the switch to another window. The skill it asks is the same as on any route: saying clearly what the model should contain, then checking what the AI added.
• You need central governance. With the Advanced license connected to Kernaro AI Hub, access runs through SSO with roles, named seats, shared agents and an audit trail. Claude with AI Power Tools acts under the architect’s own EA login and has no central audit layer of its own.
• Checks that run when the model changes. Mapping agents to EA broadcast events is a Kernaro Assist capability, so a team can have a check run when someone edits the model. Deciding which checks matter is the architect’s part.
• Staying inside Sparx Systems’ own products, with Kernaro AI Hub for stakeholder access at scale.

Jobs where Claude with AI Power Tools fits

• Analysis and build work beyond day-to-day modeling. With AI Power Tools, an architect can trace what depends on a system, compare the model against a baseline, build and check an MDG and its rulesets, generate navigable diagrams across a model, and build the reporting database that feeds Power BI. The skills give the AI a written method for each, and the architect still decides what the result should be.
• Write work without the Advanced Kernaro license. The Basic Kernaro license is read-only, and changing the model needs the Advanced license. AI Power Tools is free, carries no license term about repository type and works through the architect’s own EA session.
• Teams that already use Claude. There is no new license to buy, and the skills give the AI a method that a blank chat window does not have.

Where each one stops

A few limits the lists above leave out. For Power BI and Microsoft 365 Copilot, the Kernaro route runs through Kernaro AI Hub and its Power BI connector, not through Assist itself. AI Power Tools runs on Windows only, with EA on the same machine, in an organization that permits Claude. The reporting database it builds can sit in your own repository database, where you read it live, or in a platform such as Snowflake or Microsoft Fabric, where it refreshes on the schedule you define.

Neither one substitutes for the architect. Both produce confident output and both need someone who knows the estate to check it.

What this comparison leaves out

If your real question is cross-tool analysis, such as impact across Jira, Confluence and the repository for people who never open EA, neither of these two is the main answer. That is the territory of Kernaro AI Hub, which is built for answers across tools and for central control.

A one-week test

Do not decide from a demo. Give both tools the same three jobs on a copy of your own repository, and judge whether the output is correct before you look at how fast it came.

1. Turn a real inventory, a spreadsheet of applications perhaps, into stereotyped, connected elements in your own MDG.
2. Answer an impact question you already know the answer to, such as what depends on a particular system.
3. Run a conformance check against a standard your team uses, and have someone who knows the model read the findings.

Count the cases where the output looked right and was not. That number tells you more than the time it took, and it tells you how much review each tool will need from your architects.

A practical rule

If the job is giving modelers AI help inside EA, with governance that scales, start with Assist. If the job is the widest range of skilled work on the live model, in an organization that permits Claude, start with Claude with AI Power Tools. Many teams will use both: Assist for daily modeling, and AI Power Tools for an MDG project or a reporting build. The full comparison covers the other four options.

Questions people ask