---
source_url: https://arxiv.org/abs/2605.04816
ingested: 2026-05-09
sha256: aaa2efc51b63fbcbbf13865001d237b4c5612409d7298327a662b318f8da4f2f
---

# Building AI Companions that Prioritise Learning over Performance

**Authors:** Hassan Khosravi, Dragan Gasevic, Shazia Sadiq, Lixiang Yan, Jason Lodge, Jason Tangen, Paul Denny, Kristen DiCerbo, Simon Buckingham Shum, Ryan S. Baker
**Published:** 2026-05-06
**Venue:** arXiv
**URL:** https://arxiv.org/abs/2605.04816

## Abstract
LLMs boost short-term task performance but risk undermining genuine learning — cognitive growth, knowledge transfer, and metacognitive development. This paper introduces AI learning companions: adaptive, pedagogically informed, LLM-powered agents. Proposes a three-foundation framework: pedagogical (how students learn with AI), adaptive (how AI learns about students), and responsible design (transparent, accountable, inclusive, secure). Validated through five case studies across diverse educational contexts.