What if an SAP architecture diagram was not the end product — but the beginning of an executable architecture model?
I started this experiment with a relatively simple question: could OpenAI Codex + MCP + draw.io turn authoritative SAP documentation into a technically grounded, fully editable architecture diagram?
Three articles later, the experiment has moved considerably further.
Part 1 — From Documentation to Editable Architecture
Can an AI agent read SAP source material, respect SAP’s architecture vocabulary and construct a real editable diagram rather than generate another attractive but flattened image?
Part 2 — Building It for Real
I document the actual Windows setup with Codex CLI, MCP, ARC-DRAW and draw.io—including SAP icon validation, evidence-based connectors, pre-draw architecture audits, deterministic geometry and the lessons from diagrams that were technically correct but visually mediocre.
Part 3 — From Editable Diagram to Architecture-as-Data
This is where the idea becomes much more interesting. What if systems, interfaces, requirements, tests, architecture decisions, operational signals and evidence become connected objects that both architects and controlled AI agents can reason over?
That leads naturally toward SAP Cloud ALM, observability, architecture governance and autonomous ALM—but with an important principle:
Not autonomy first.
Context first.
Evidence first.
Governance first.
Then automate what can safely be automated.
The series is ultimately not about getting AI to draw better boxes.
It is about moving from documentation → architecture → lifecycle evidence → AI-assisted reasoning → governed action.
Cover concept
I created the overall cover around exactly that progression: Part 1 = knowledge to editable architecture; Part 2 = the engineering/toolchain; Part 3 = architecture-as-data and governed AI.
The headline “From Documentation to Decisions” is deliberately broader than “AI-generated diagrams,” because Part 3 establishes that the editable diagram is an entry point into structured, traceable and potentially executable architecture knowledge.
Read More Technology Blog Posts by Members articles
#abap