AI code generators

Microsoft has published MXC, an open-source sandboxed code execution system, on GitHub. The tool is designed to run untrusted code in isolated environments, a growing need as AI applications increasingly execute model-generated code. Developers are discussing the project's approach to sandboxing and its potential uses in building safer application infrastructure.

JetBrains has published a developer diary walking through how it built a RAG (retrieval-augmented generation) pipeline for semantic code search, sharing field notes and lessons learned along the way. The post is drawing attention among developers interested in how AI-powered code search can be engineered in practice, including the trade-offs and pitfalls of retrieval systems applied to large codebases.

A new study reported by Ars Technica finds that AI coding agents do generate more code, but teams do not ship more software, because the gains are absorbed by a human review bottleneck. The findings suggest that developer time spent reviewing and validating machine-written code limits overall productivity improvements.

A new study reported by Ars Technica finds that AI coding agents dramatically increase the amount of code generated, yet overall software output does not improve. The efficiency gains are reportedly absorbed by a human review bottleneck, as developers must still spend time evaluating, testing and integrating the machine-written code. The findings challenge assumptions that AI coding tools automatically translate into faster software delivery and are prompting debate about where the real limits of AI productivity lie.

AI coding assistants are complicating software bill of materials (SBOM) practices, as generated code and dependencies often fall outside documented component inventories. The concern lands just as the EU Cyber Resilience Act approaches, a law that will require manufacturers to maintain accurate SBOMs and manage security risks across products. Security teams are being urged to rethink how they track AI-generated dependencies before compliance deadlines arrive.

A tool called REA, or Reverse Engineer Anything, connects AI coding agents to professional reverse-engineering platforms Ghidra and IDA. It lets researchers inspect software, trace behavior and reconstruct features while generating supporting evidence for their findings. The project is drawing attention in the cybersecurity community, where analysts see potential for automating parts of binary analysis and malware research, though some are weighing the risks of handing such work to AI agents.

A new study covered by Ars Technica finds that AI coding agents dramatically increase the amount of code generated, yet overall software output does not grow. The efficiency gains are instead absorbed by a human review bottleneck, as developers must still read, verify, and integrate what the agents produce. The finding challenges assumptions that AI coding tools automatically translate into faster software delivery and is fueling debate about where real productivity limits lie.

Analysis by Ars Technica argues that AI coding agents are driving a surge in the volume of code being written, but that this does not translate into more completed, working software. The piece suggests that generating code faster is not the same as shipping finished products, a distinction critics say is often lost in debates over AI's impact on software engineering productivity.

A new preprint examines whether code written by AI coding agents is worse than human-written code, and finds no single number answers the question. According to the paper, pull requests submitted by AI agents get reverted at rates of 6.1% or 14.5%, depending on which vendor's tool produced the code. The findings suggest code quality from agentic tools differs significantly across providers, complicating sweeping claims about AI-generated software.