# Not Everything Needs an LLM

- HackerByte page: <https://hackerbyte.net/en/stories/an-evaluation-of-the-unnecessary-use-of-large-language-models-llms-for-solving-e-50535740>
- Original source: <https://newsletter.posthog.com/p/decision-models>
- Hacker News discussion: <https://news.ycombinator.com/item?id=50535740>
- Submitted to Hacker News: 2026-11-17T18:16:18.000Z

An evaluation of the unnecessary use of large language models \(LLMs\) for solving every problem.

## AI-generated summary

This article argues that Large Language Models \(LLMs\) are often overkill for simpler tasks and proposes using decision models as a more efficient alternative. Decision models provide structured, probabilistic outputs, making model uncertainty explicit and offering significant cost and speed benefits over LLMs. PostHog has successfully integrated decision models into production for tasks like PII scrubbing and as a model router, demonstrating their utility in filtering tasks before an LLM call.

## Key points

- Decision models offer structured, probabilistic outputs unlike LLM's free\-form prose.
- They are significantly faster and cheaper than frontier LLMs.
- Used at PostHog for PII scrubbing in session recordings, reducing LLM calls by 98.7%.
- Effective for routing LLM requests to the cheapest, most suitable model.
- Best for tasks too complex for regex but too simple for full LLMs.
