CLAUDE CODE AGENTS: FUNDAMENTALS - Chapter 10, Exercise 1
Why Model Choice Is a Genuine Trade-Off for a Research Agent
====================================================================================

QUESTION:
Explain why a research agent's ideal model choice is described in this
chapter as a genuine trade-off, unlike most other agent types in this
course.

SOLUTION / EXPLANATION:

For most agents earlier in this course, model choice leaned consistently
toward "use the most capable model available," because their jobs involved
real judgment where mistakes are costly: a code-review agent catching
subtle bugs, a coding agent implementing correct behavior, a teaching agent
adapting an explanation well. In each of those cases, a more capable model
was described as straightforwardly better suited to the task's actual
difficulty.

A research agent's job is different in a way that makes the answer
genuinely depend on the specific task rather than pointing consistently in
one direction. Some research tasks are simple, narrow lookups - "which file
defines this one function" - where a faster, lighter model can find the
answer just as reliably and much more quickly, since the task doesn't
require deep reasoning, only straightforward pattern matching. Other
research tasks are genuinely wide and ambiguous - searching an unfamiliar
codebase where the right search terms aren't obvious upfront, requiring the
agent to reason about where to look next based on what earlier searches
turned up. That kind of adaptive, uncertain search benefits genuinely from
a more capable model's better judgment.

So unlike Chapter 6's review agent or Chapter 5's coding agent, where "most
capable available" was the straightforward answer because the job always
demands real judgment, a research agent's ideal model genuinely depends on
which of these two very different kinds of task it's actually being asked
to do - making it a real trade-off to weigh case by case, not a fixed
default.

--------------------------------------------------------------------------
WHY THIS WORKS AS AN ANSWER:
It contrasts the consistent "most capable model" guidance given for other
agent types against the genuinely task-dependent answer for a research
agent, explaining specifically why simple lookups and wide ambiguous
searches call for different model choices rather than one answer fitting
both.