Field notes on enterprise AI, agents, systems, and the way software changes organizations.
I keep coming back to the same question: as language models start interacting with tools, memory, users, and real enterprise constraints, what does the work around them actually look like? This site is where I think it through in the open. Less a personal brand, more a notebook that happens to be public.
Start here
- Agent architecture map Map for concrete architecture nodes: MCP, tool calling, permission boundaries, traces, orchestration, and state.
- Enterprise AI map Map for governance, write access, systems of record, and approval boundaries.
- Governance map Map for control and accountability: permission boundaries, approvals, audit trails, and rollback.
- Interfaces map Map for the action interfaces a model uses: tool calling, tool schemas, MCP, and workspace state.
- Memory map Map for memory and state: federated memory, the promotion pipeline, workspace state, and traces.
- From chatbots to system operators Working-theory essay: LLM agents are turning into scoped operators of real systems.
- Concept notes Reusable concept pages for concrete nodes like MCP, tool schema, approval gate, rollback, and workspace state.
- Write access in regulated environments Field note: the blocker isn't the regulator, it's that orgs don't know how yet and the APIs were never built as an agent's surface.
- How this works The living wiki idea: version controlled, interlinked, an AI maintains it but I edit and approve every change.
Everything here, content and code, lives in one public repository: github.com/FVossebeld/FVossebeld.github.io. Search is top left, and each page's connections (its backlinks and a small graph) sit at the foot. Pages link to each other, so follow the threads.