Trajectory-Informed Memory Generation for Self-Improving Agent Systems Conference: Research done at IBM. Published; Published Feb 2026; Yorktown Heights, NY Technical Report describing research done at IBM CCS: Computing methodologies Information extraction CCS: Computing methodologies Multi-agent systems CCS: Computing methodologies Knowledge representation and reasoning CCS: Information systems Enterprise applications CCS: Information systems Information retrieval
Gaodan Fang, Vatche Isahagian, K. R. Jayaram, Ritesh Kumar, Vinod Muthusamy, Punleuk Oum, Gegi Thomas · 2026
Read Trajectory-Informed Memory Generation for Self-Improving Agent Systems Conference: Research done at IBM. Published; Published Feb 2026; Yorktown Heights, NY Technical Report describing research done at IBM CCS: Computing methodologies Information extraction CCS: Computing methodologies Multi-agent systems CCS: Computing methodologies Knowledge representation and reasoning CCS: Information systems Enterprise applications CCS: Information systems Information retrieval by Gaodan Fang, Vatche Isahagian, K. R. Jayaram, Ritesh Kumar, Vinod Muthusamy, Punleuk Oum, Gegi Thomas (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-10600.txt.
Description
Gaodan Fang and co-authors present a framework that turns agent execution traces into memory an agent can reuse. Four parts do the work. An extractor reads reasoning patterns, and an analyzer pins down which decisions caused failures or waste. A generator writes strategy, recovery, and optimization tips, and a retrieval system injects the right tips into later prompts. Unlike memory that stores plain chat facts, this one keeps structured lessons with provenance. On AppWorld, held-out task completion rose notably, with the biggest gains on complex tasks.
Catalog record
| CREATOR | Gaodan Fang, Vatche Isahagian, K. R. Jayaram, Ritesh Kumar, Vinod Muthusamy, Punleuk Oum, Gegi Thomas |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 75487 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.10600