Already 10 months old when the scout caught it.
Context plumbing
What it is
Context plumbing is the engineering discipline of moving context — the facts, state, and signals an agent needs — from wherever it's produced to wherever an agent is actually running, at the moment it's needed. Coined by Matt Webb, it frames an AI system's technical architecture as a set of pipes connecting context "sources" (a changed document, a new email, a shifted calendar, an updated tool list) to context "sinks" (the point of user intent, which is often not where the model itself executes).
Why it's emerging now
As agents move from single-shot chat into longer-running, multi-tool, multi-surface workflows, the hard problem has shifted from writing a good prompt to routing the right facts to the right place before they go stale. Context is scattered across disparate, independently-updating sources, and the agent's execution point is rarely colocated with all of them — so getting an agent to act well increasingly depends on the plumbing that ferries context between sources and sinks, not just on how the prompt is worded once it arrives.
How it relates to what you know
It's a sibling to context engineering already on the ledger, but a narrower one: context engineering is about deciding what belongs in the context window for a given task, while context plumbing is about the infrastructure that keeps that window fed from live, distributed sources as they change. It belongs to the same family of "-engineering" successor framings as loop engineering and graph engineering — each takes one piece of agent-building (routing data, structuring control flow, structuring execution topology) and names it as its own discipline.