How AI agents are being used to fix real-world software problems

A survey of 242 studies maps the tasks, training methods and tests behind automated software issue resolution.

Editorial illustration — not from the study.

Software issue resolution involves turning a user’s description of a problem into changes to a software project. The survey reviews 242 recent studies on AI agents designed to carry out this work, including the repeated exploration, reasoning and feedback needed for complex issues.

The authors organize the research around three areas: benchmarks used to test systems, techniques used to build them, and empirical studies that evaluate them. They also describe open challenges and possible directions for future research, including the growing role of reinforcement learning in training these systems.

What the survey maps

The survey describes a general workflow for AI-based software issue resolution and proposes a taxonomy covering three dimensions: benchmarks, techniques and empirical studies. It finds that reinforcement learning has become an increasingly important training approach for agentic systems in software engineering. The review also identifies unresolved challenges and outlines directions for future research.

Why software fixes matter

Fixing software issues is a central part of software maintenance, but real-world problems often require multiple rounds of investigation, planning, code changes and feedback. By organizing a rapidly expanding body of research, the survey shows how AI agents are being developed for these longer, more involved tasks. The authors also present software issue resolution as a practical setting for evaluating an AI system’s reasoning, planning and ability to act, linking developments in artificial intelligence with software engineering.

Evidence and limits

This is a systematic survey of 242 recent studies, not a new experiment testing one software-resolution system. Its conclusions summarize the research covered by the review, so they do not by themselves establish that AI agents reliably resolve software issues in practice. The abstract does not provide comparative performance results or details about the surveyed studies’ individual limitations.

Programmable hardware speeds matrix calculations up to 6.3 times

Researchers tested a programmable hardware design that imitates selected quantum-computing ideas to run matrix calculations in parallel. In representative benchmarks, the design reached up to 6.3 times the speed and 4.1 times the energy efficiency of conventional matrix co-processors and embedded graphics processors.

2 days ago·13 views

AI & Computing

AI and docking agreed in some drug screens, but not all

Researchers evaluated three AI systems on hundreds of molecules bound to a SARS-CoV-2 protein and in three prospective drug screens. The results suggest that combining AI-based predictions with traditional docking may help prioritize candidates, although no single approach worked best in every test.

4 days ago·50 views

AI & Computing

A hardware monitor detected simulated attacks on car networks in 14 microseconds

Researchers developed a hardware-assisted system intended to detect message injection, spoofing, flooding and impersonation attacks on automotive networks. In tests using simulated traffic and a prototype automotive processor system, it reported an average detection time of 14 microseconds and less than 2 percent CPU overhead.

5 days ago·42 views