---
title: "Comment by Vlad Zoff on \"Your AI Agent Improved Its Score. Did It Improve the System?\" | Hashnode"
description: "Benchmark improvement can be a dangerous proxy. An agent can get better at the metric by changing the system around the task, while the actual workflow becomes"
url: "https://hashnode.com/posts/your-ai-agent-improved-its-score-did-it-improve-the-system/6aa5d1b67cfde2f47ac4f72d/comment/6abdf372d213d3267f0d0175"
---

[Discussion](https://hashnode.com/posts/your-ai-agent-improved-its-score-did-it-improve-the-system/6aa5d1b67cfde2f47ac4f72d)

Comment [Your AI Agent Improved Its Score. Did It Improve the System?](https://hashnode.com/posts/your-ai-agent-improved-its-score-did-it-improve-the-system/6aa5d1b67cfde2f47ac4f72d)

[Vlad Zoff](https://hashnode.com/@vladzoff)

[Oct 1](https://hashnode.com/posts/your-ai-agent-improved-its-score-did-it-improve-the-system/6aa5d1b67cfde2f47ac4f72d/comment/6abdf372d213d3267f0d0175)

Benchmark improvement can be a dangerous proxy. An agent can get better at the metric by changing the system around the task, while the actual workflow becomes harder to reason about. I like separating the task outcome, the evaluation signal, and the system changes instead of treating one score as the whole result.
