LLMs, GenAI, RAG, agents, evaluation · insight-0643

Prove one agent is insufficient before splitting work across several agents

Situation
A design proposes separate planner, researcher, critic and executor agents for a task one agent has not yet attempted.
Action
Build and evaluate the single-agent (or workflow) version first; add another agent only for a named failure it fixes, and compare both on the same task set, cost and latency.
Reason
Anthropic's guidance advises adding agentic complexity only when it demonstrably improves outcomes; every extra model call is another place for errors to compound.
Boundary
Architecture reviews for GenAI features (days 55 and 58).

Symptom or risk

Each hand-off loses context and adds calls, so cost, latency and failure points multiply while the benefit is unmeasured.

Hard isolation needs, such as a quarantined model that reads untrusted data without tool access, justify separation for security rather than quality.

Specialised agents can use smaller models per role, which may lower cost; measure it.

Evaluating the multi-agent system only on showcase tasks.

Anthropic's guidance recommends finding the simplest solution and increasing complexity only when needed, noting that agentic systems trade latency and cost for task performance.
Building Effective Agents