AI / Agentic AI Interview Questions II
1. Which is better and why: single powerful agent vs multiple specialized agents for a complex task?
| Single powerful agent | Multiple specialized agents |
| Simpler to build, one set of instructions and tools | More setup complexity, coordination between agents required |
| Can lose focus juggling many unrelated responsibilities at once | Each agent stays focused on a narrower, well-defined responsibility |
| One context window shared across the entire task | Each agent's context stays smaller and more relevant to its specific role |
| Slower for tasks that could otherwise run parts in parallel | Can run independent sub-tasks concurrently for speed |
Neither option is universally better. A single agent is often the right starting point for a genuinely simple, linear task, since it avoids the coordination overhead of a multi-agent system entirely. As task complexity grows, particularly when sub-tasks are independent or require genuinely different expertise, splitting into specialized agents tends to produce a faster, more maintainable, and more focused system, at the cost of added orchestration complexity that has to be managed deliberately.
2. Explain the internal working of MCP's client-server orchestration model?
- The MCP client, which hosts the LLM runtime and agent reasoning logic, connects to one or more MCP servers
- It requests the list of available tools, and each server responds with its tool schemas, names, descriptions, and expected parameters
- The agent's planner selects a tool based purely on that returned name and description, without inspecting how the tool is actually implemented on the server side
- The client sends a tool call to the server using JSON-RPC, and because MCP supports long-lived, stateful sessions, results can stream back progressively rather than arriving only once at the very end
- The server itself typically remains stateless and reusable, with the actual decision-making and session logic living in the client
This design keeps servers simple, swappable providers of capability, while concentrating orchestration intelligence in the client, which is also why production deployments increasingly route multiple MCP servers through a centralized gateway to manage authentication, rate limiting, and tool exposure consistently.
