Architectural Governance & Bounded Autonomy for Enterprise Agentic AI
Articles on agentic AI hosted by Vincent Caldeira and written with collaborating authors, published as part of the Agentic AI Foundation Ambassador program — technical writing on enterprise agent architecture, safety, and open-source interoperability.
Topics
Projects
Open decision models as the agent control plane: typed Choice and Score beat JSON chat for routing, pair with proven search for passages, and gate MCP tools with structured probabilities.
An 8B model with a short tool list can beat a 70B model on the full catalog. Keep the cheaper model; filter what it sees — no fine-tune required.
From vibe checks to Evaluation-Driven Development: MLflow tracing, MCP tool-call evaluation, and MemAlign CLHF for enterprise financial agents.
AGENTS.md should orchestrate specs, not hoard them. How spec-driven teams govern multi-agent context — and stop catastrophic remembering from bloating the harness.
An agent will do exactly what you tell it to do — and every paragraph you add to the harness costs attention, accuracy, and GPU memory.
How MCP 2026-07-28 Multi Round-Trip Requests let enterprise agents pause for human approval without long-lived server connections — with a LangGraph and agentgateway design walkthrough.
Day two of enterprise agentic engineering — distributing all four SSDF pillars (AGENTS.md personas, CodeGuard, Spec-Kit templates, Foundry CI context) through Lola modules and a single .lola-req manifest.
A practical base template strategy for Python agent development — combining AGENTS.md, GitHub Spec-Kit, and OGX into a governed spec-driven workflow.
A conceptual overview of four foundational policy layers — perimeter, runtime, metacognitive, and systemic — for governing enterprise agent run loops with agentgateway and MCP.