March 4, 2026 https://classroom.github.com/a/wr5tG_By
Links
When to Use LLMs
Before diving into agents, RAG, and workflows — the most important skill is knowing when to use LLMs and when not to.
Good Fits for LLMs
- Text summarization and transformation: Condense documents while preserving key information
- Structured data extraction: Convert unstructured text to structured formats (JSON, tables)
- Content classification: Categorize by type, topic, sentiment
- Question answering over documents: Answer questions based on provided context
- Draft generation with review: First drafts that humans refine
Poor Fits for LLMs
Conversely, some tasks look like they should work but consistently produce poor results:
- Precise calculations: Use tools (calculators, code) instead
- Factual retrieval without verification: LLMs may hallucinate
- Real-time data without external connection: Models have knowledge cutoffs
- High-stakes autonomous decisions: Require human oversight
- Deterministic logic: Use rule engines instead
Reference Card: LLM Decision Framework
| Question |
Yes → |
No → |
| Can you describe the task clearly? |
Good candidate |
Clarify requirements first |
| Are errors catchable? |
Proceed with validation |
Add human review or avoid |
| Can you validate outputs? |
Automate with checks |
Use expert oversight |
| Do you have domain expertise to evaluate? |
LLM amplifies your skill |
Risk of undetected errors |
Common Failure Modes