AI & ML Paradigm Shift

Atlas introduces 'Compiled Memory,' which rewrites an agent's system prompt with distilled task experience rather than using RAG or fine-tuning.

March 18, 2026

Original Paper

Compiled Memory: Not More Information, but More Precise Instructions for Language Agents

James Rhodes, George Kang

arXiv · 2603.15666

The Takeaway

It treats memory as instruction evolution rather than information retrieval, significantly improving precision on complex tasks like legal analysis (+12.5pp) without the latency or context-window costs of RAG.

From the abstract

Existing memory systems for language agents address memory management: how to retrieve and page more information within a context budget. We address a complementary problem -- memory utility: what experience is worth keeping, and how it should change agent behavior. We present Atlas, a memory kernel that compiles accumulated task experience into an agent's instruction structure -- without fine-tuning, RAG, or human intervention. Memory is distillation, not storage; delivery is instruction rewrit