Live Software Adaptation
Methods and benchmarks for LLM agents that modify software behavior or code at runtime in response to observed failures or user needs.
5 papers
Where this stands
The written synthesis of this thread is for subscribers. Subscribe.
How this thread developed
Sept 2026 · University of Kentucky, University College London
Embedded AI agents can adapt software during live use
Demonstrates agents that modify software logic and interfaces at runtime, opening a new direction for self-adaptive systems.
2 further papers
Sept 2026
Multi-image evidence can help models repair software, but unreliably
Introduces a multi-image software-repair benchmark and evaluates how visual access affects success rates, showing unreliability.
Oct 2026 · University of California, Santa Barbara, Microsoft
Training models on runtime program-state reasoning boosts automated software engineering
Introduces program-state reasoning tasks and a staged post-training pipeline to build Comet-9B, enhancing automated software engineering capabilities relevant for live software adaptation.
3 of 5 papers shown