Choosing Between One Big Agent and Five Specialists
One generalist with all tools is simpler. Specialists are more reliable. The decision rule, with cost numbers, that picks the right shape per use case.
The rule
Specialists win when MTTR matters and the failure modes are diverse. Generalists win when scope is narrow and integration cost dominates.
Specialists also win when each specialty has its own eval suite and prompt-engineering owner. Without dedicated owners, specialists rot.
Most production SRE work is diverse failure modes; specialists are usually right. Resist the simplification of "one big agent."
Cost numbers
Generalist: 1 prompt, ~8k tokens. ~$0.04 per run. Higher latency due to bigger context.
Specialists: 5 prompts × 2k tokens each. ~$0.05 per run. Faster per specialist; orchestration overhead adds latency.
The cost is similar. The reliability difference is large: specialists tend to be more reliable on their domain because their prompts are tighter.
When generalists win
Single-domain agents (just DNS, just K8s, just databases). Within a single domain, a generalist is fine; specialisation does not help.
Low-volume agents (few runs per day). The orchestration overhead of specialists is not amortised; the simpler shape wins.
Early-stage agents. Ship a generalist first; split into specialists after observing failure modes.
How to split into specialists
Split by failure mode, not by tool. "Database specialist" handles all database failures, not just "queries" or "connections."
Each specialist gets its own prompt, its own tools, its own eval suite. Treat them as independent products.
Orchestration is code, not a meta-agent. The orchestrator dispatches based on input; the dispatch is deterministic.
When to rejoin specialists into a generalist
If two specialists are converging on similar prompts and similar failures, consider rejoining. Maintaining duplicate effort is waste.
If usage of one specialist is low (under 5 runs per week), rejoin into a sibling. The specialist is not paying its keep.
Rejoin is rare. Most evolution is the other direction; specialists deepen rather than rejoin.