Concept guide · Agents, tools, and MCP
When should you not build an agent?
A slower explanation for when the short interview answer is not enough. Start with the intuition, then connect it to the technical idea and the way an interviewer expects you to reason about it.
ML area
Generative AI → Agent boundaries
Interview format
Core · Scenario
Question
GENAI-076
Start with the idea
In simple terms
Do not build an agent when a small deterministic workflow can solve the problem. Agents add decisions, latency, cost, and failure modes, so the extra autonomy should solve a real uncertainty in the task.
Make it concrete
A password-reset flow should be a fixed, permissioned workflow. A research assistant that must choose which sources and follow-up searches to use may justify an agent loop.
Build the answer
How to approach it
- 1Write the known steps and failure policy before choosing an agent.
- 2Identify the specific uncertainty that requires model-driven routing.
- 3Keep sensitive actions behind fixed checks, approvals, and least-privilege tools.
Go one level deeper
The technical explanation
Start with the simplest baseline: direct model call, retrieval pipeline, or explicit state machine. Add an agent when the number or order of steps cannot be known in advance and the value of flexible planning exceeds the operational cost. Avoid agents for irreversible high-impact actions unless approval and deterministic validation remain in control.
Say this in the interview
Do not use an agent when the steps are known, the task is safety-critical and deterministic control is required, or a simple retrieval or workflow is sufficient. Agents add latency, cost, nondeterminism, and debugging surface. Start with a fixed chain and add autonomy only when the measured problem requires it.
Keep preparing