MAPLE: A Sub-Agent Architecture for Memory, Learning, and Personalization in Agentic AI Systems Conference: Submitted to the Adaptive and Learning Agents Workshop (ALA 2026) at AAMAS 2026; May 25 – 26, 2026; Paphos, Cyprus
Deepak Babu Piskala · 2026
Read MAPLE: A Sub-Agent Architecture for Memory, Learning, and Personalization in Agentic AI Systems Conference: Submitted to the Adaptive and Learning Agents Workshop (ALA 2026) at AAMAS 2026; May 25 – 26, 2026; Paphos, Cyprus by Deepak Babu Piskala (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-2602-13258.txt.
Description
AI agents that serve one person well still tend to forget, relearn from scratch, or blur distinct jobs into one muddled memory. MAPLE argues that memory, learning, and personalization are three separate mechanisms, each needing its own tools, pace, and tuning, and splits them into dedicated sub-agents. Memory stores and fetches past interactions, learning draws lessons from them in the background, and personalization applies those lessons live within tight context limits.
On the MAPLE-Personas benchmark, the split design lifted personalization scores by 14.6 percent over a memory-free baseline and raised trait use from 45 percent to 75 percent of cases. The gain suggests agents can adapt to a user in a lasting way only when each job gets its own machinery. Deepak Babu Piskala is the sole author of the paper.
Catalog record
| CREATOR | Deepak Babu Piskala |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-13258