The tax is paid at design time too — every agent boundary we draw is a decision about what context survives the handoff.
watching → emerging → established
Seen across products but the shape is still settling. Adopt with an exit: expect the anatomy to shift.
8 documented sightings since Aug 2026 — every one traces to a dated source.
See the evidence →Martin Fowler's piece reframes the appeal of multi-agent decomposition — breaking a big task into specialized sub-agents — as a budget problem rather than a pure architecture win. Each handoff between orchestrator and sub-agent requires reconstructing context: the orchestrator must package what the sub-agent needs to know, and the sub-agent must interpret it correctly before doing any real work. That overhead is the "orchestrator's tax," and it compounds when chains get deep.
The argument cuts against the reflex to decompose everything into micro-agents for modularity's sake. Fowler's position is that the context reconstruction cost is often underestimated at design time, and that a single well-scoped agent holding more context in one working memory may outperform a delegation chain that hemorrhages fidelity at each junction. The evidence is primarily argumentative — no benchmarks — but the conceptual frame is precise enough to be useful.
For UX and systems designers, the practical implication is that every agent-to-agent delegation boundary is also a design problem: what context travels with the handoff, what gets lost, and how does a human recover when a downstream sub-agent acted on a thin or stale picture of the task. Legibility of those boundaries — who handed what to whom, with what context attached — becomes a first-order concern rather than a logging afterthought.
The summary above is generated; the note at the top is the editorial judgment. Primary source ↗
Agentic experience design, coding agents, MCP, and the signals that matter — weekly, free, in about five minutes.
Free. No spam. Unsubscribe anytime.