---
title: "Evo-MedAgent Teaches Frozen AI Doctors to Learn From Past Mistakes"
subtitle: "A training-free memory module lets frozen chest X-ray agents accumulate lessons across cases, lifting MCQ accuracy up to 11 points without any fine-tuning."
url: "https://alphasignal.ai/news/evo-medagent-teaches-frozen-ai-doctors-to-learn-from-past-mistakes"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-04-15T00:00:00.000Z"
updated: "2026-08-23T20:41:24.056Z"
type: "Paper"
topics:
  - "Agents"
  - "Benchmarks"
subtopics:
  - "Memory"
  - "Tool Use"
upvotes: 921
source_url: "https://arxiv.org/abs/2604.14475"
comments: 0
truncated: true
---

# Evo-MedAgent Teaches Frozen AI Doctors to Learn From Past Mistakes

> A training-free memory module lets frozen chest X-ray agents accumulate lessons across cases, lifting MCQ accuracy up to 11 points without any fine-tuning.

A radiologist who misreads a subtle pneumothorax on Monday is unlikely to make the same mistake by Friday. Today's tool-augmented medical agents don't have that luxury. They can call classifiers, segmenters, and vision-language models in sequence, but every patient arrives as a blank slate. A new paper called [Evo-MedAgent](https://arxiv.org/abs/2604.14475) tries to close that gap with a plug-in memory system that lets a frozen model learn from its own mistakes at test time.

The headline numbers are unusually clean for an agent paper. On ChestAgentBench, Evo-MedAgent raises multiple-choice accuracy from 0.68 to 0.79 on GPT-5-mini, and from 0.76 to 0.87 on Gemini-3 Flash. No fine-tuning, no reinforcement learning, no gradient updates anywhere in the stack.

## Why static agents keep tripping on the same rock

Recent systems like [MedRAX](https://bowang-lab.github.io/MedRAX/), CheXagent, and MDAgents orchestrate specialist tools through a ReAct-style reasoning loop. They work, but they forget. Each case is solved in isolation, experience never accumulates, and recurrent reasoning mistakes cannot be corrected without expensive reinforcement learning.

The authors argue the errors are often procedural rather than perceptual: wrong tool ordering, badly chosen hyperparameters, over-trust in a noisy classifier. Retraining the underlying VLM to fix any single one of these is absurdly expensive, and there is no obvious way to encode a lesson like "when you see a unilateral effusion in a teenager, also check the ribs for a spur."

## Three memory stores, each doing a different job

Evo-MedAgent factors memory into three components that get read before inference and written after ground-truth feedback:

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [Evo-MedAgent: Beyond One-Shot Diagnosis with Agents That Remember, Reflect, and Improve](https://arxiv.org/html/2604.14475)
- [MedRAX: Medical Reasoning Agent for Chest X-ray](https://bowang-lab.github.io/MedRAX/)
- [ChestAgentBench dataset](https://huggingface.co/datasets/wanglab/chest-agent-bench)
- [MedRAX GitHub repository](https://github.com/bowang-lab/medrax)