Can AI answer any question about my enterprise architecture?

Institutional memory is still mostly people

Ask a hard question in most enterprises — “what breaks if we retire this platform?”, “who owns this integration really?”, “why was this pattern chosen?” — and the answer path is social. Find the right person. Hope they are free. Hope their mental model is current.

Documents help until they do not: search returns ten near-matches, none authoritative. Repositories help until attributes are too thin to explain the decision. The continuum fails at the moment of interrogation.

Why “any question” needs an Architecture Decision Workspace

A generic chatbot over a document dump will invent confidently. A catalogue search will under-answer. DecisionPen’s enterprise Ask is built for a third path:

• Map-grounded — answers reference structured entities and relationships, not only free text.
• Evidence-aware — claims can point back to documents and dialogue that justify them.
• Metamodel-aware — the agent understands architectural categories and valid relationships, reducing category errors.
• Honest about gaps — unknown is preferable to fabricated certainty when the workspace has not yet captured the fact.

Walk through: how DecisionPen supports discovery and decision questions on a curated map

Here is the practical loop teams use when Ask sits on a living landscape rather than a pile of documents:

1. Start from the curated map — applications, capabilities, technologies, and integrations already proposed and confirmed, with evidence still attached from HLDs, transcripts, and other sources.
2. Ask a discovery question — for example, “What depends on Customer Payments Hub?”, “Which capabilities does this application support?”, or “Where is ownership unclear?”
3. Read a map-grounded answer — DecisionPen traverses relationships and returns entities in context, not a generic essay assembled from similar paragraphs.
4. Follow the evidence — open the cited passages and dialogue that justify each claim so architects can trust or challenge the answer.
5. Ask a decision question — retire, invest, sequence a migration, or scope a programme: “What is the blast radius if we retire the legacy ledger?” The same curated map powers the answer.
6. Close gaps deliberately — when the workspace does not yet know, capture the missing context (document, meeting, or natural-language update), curate it, and ask again — so institutional memory improves with every programme.

Interrogate the living landscape

Walk through how DecisionPen supports discovery and decision questions on a curated map — Customer Payments Hub onboarding.

Ready to try it yourself?

See it live

From people’s heads to a shared decision system

Moving institutional memory into DecisionPen is not a single migration event. It is the cumulative effect of the modelling loop:

1. Capture what meetings and documents already contain
2. Curate what becomes canonical structure
3. Keep evidence attached so answers remain explainable
4. Let every new project add to the same memory
5. Ask questions against that shared base instead of restarting discovery

Over time, the organisation’s scarce experts stop being the only index of the landscape. They become governors and decision-makers — which is what they were hired for.

Continue from here

If you are evaluating ChatGPT, Copilot, or a Confluence RAG bot as a substitute, read why plain GenAI cannot match a metamodel-aware EA agent. Feature deep-dives will continue to land in this Docs section as they are ready.