Essays and field essays
Strategic essays and field essays from my personal architecture work. Some pieces advance the longer thesis arc; others externalize practical signals that are becoming reusable patterns for operating in the AI era.
Once AI usage spreads across teams, models, and action types, economics stops being a finance dashboard problem. It becomes an architecture problem where routing, budget envelopes, quality thresholds, and security controls have to shape runtime behavior.
Patterns referenced: Cheap-first routing envelope · Showback stabilizes dashboards, not spend · Security control profile by action tier
Security in agentic systems cannot be a single global rule set. Different actions carry different blast radius. The architecture requirement is action-tier security boundaries: policy gates, evidence requirements, and escalation rules that scale with decision impact.
Patterns referenced: Security control profile by action tier · Policy-only security without runtime evidence creates false confidence · Agent trace sink
In deterministic systems, failures produce errors. In agentic systems, failures produce confident wrong outcomes — quietly. Observability is not an infrastructure concern; it is the runtime expression of governance and the foundational requirement for enterprise agentic deployments that expect to be trusted.
Patterns referenced: Agent trace sink · Runtime governance control plane · Policy gateway
Why enterprise AI will fail if it is introduced as a patchwork of isolated agents and workflow automations. The real architectural challenge is building a decision fabric that connects context, policy, orchestration, escalation, and accountability.
Patterns referenced: Decision context envelope · Policy gateway · Runtime governance control plane
Why governance cannot stay trapped in review documents in agentic environments. It needs to be executable through architecture control planes that define policy, route decisions, and enforce safe autonomy at runtime.
Patterns referenced: Runtime governance control plane · Policy gateway · Change-absorption baseline
Most organisations treat AI adoption as a tooling conversation. The harder problem is structural: data ownership, decision boundaries, operating rhythm, and where autonomous behaviour is actually safe. Without those foundations, AI becomes noise.
Patterns referenced: Context contracts for MCP servers · Decision context envelope · Change-absorption baseline
The interesting future is not agents doing more things — it is agents doing bounded things well, inside explicit policy and governance constraints. The more autonomy increases, the more clarity matters in the surrounding system design.
Patterns referenced: Policy gateway · Runtime governance control plane · Decision context envelope
Early ideas benefit from compression: one user signal, one sharp problem, one inspectable experiment. The goal is not to scale an idea too early, but to discover whether the underlying pattern deserves to exist.
Patterns referenced: Signal-first incubation loop · Change-absorption baseline
The organisations that move fastest over time will not be the ones with the most tools. They will be the ones with cleaner system boundaries, more coherent platforms, and a better ability to absorb change without multiplying complexity.
Patterns referenced: Change-absorption baseline · Runtime governance control plane · Context contracts for MCP servers