When should you not use an LLM?

Identifies tasks better served by deterministic code, search, rules, or conventional models because of correctness, latency, cost, or control.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

I would not use an LLM when a simpler deterministic solution satisfies the requirement more reliably. Exact arithmetic, authorization, database constraints, well-defined transformations, and state transitions should normally stay in code. Plain search may be better when users need exact documents, and a conventional classifier may be better for a stable labeled task with strict latency or cost limits. I also avoid an LLM when the consequence of an unverified output is unacceptable.

Full answer below

Practice focus

Technology judgment that starts from the product problem rather than forcing an LLM into every feature.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
A mature AI design begins with the option not to use a generative model.
I would not use an LLM when a simpler deterministic solution satisfies the requirement more reliably. Exact arithmetic, authorization, database constraints, well-defined transformations, and state transitions should normally stay in code. Plain search may be better when users need exact documents, and a conventional classifier may be better for a stable labeled task with strict latency or cost limits. I also avoid an LLM when the consequence of an unverified output is unacceptable.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
The decision depends on the required behavior, not whether the input is text. An LLM is useful for ambiguous language, summarization, drafting, and flexible interpretation, but it adds variable output, data-handling questions, latency, provider dependency, and evaluation work. For an invoice total I use deterministic calculations. For explaining an invoice in natural language, a model may help after the numbers are computed and supplied as read-only facts.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Using an LLM for arithmetic or authorization because it can produce an answer.
Assuming all natural-language input requires generation.
Ignoring a search, rules, or conventional-ML baseline.
Treating an impressive demo as evidence of operational fit.

When should you not use an LLM?

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue practice

Tech stacks

Skill areas

Human editorial direction

Was this drill useful?

Found this answer useful?