AI agents can now rewrite code far faster than engineers can review it, shifting the central challenge of large-scale modernization from generating code to establishing trust in its correctness. Microsoft describes using agents, static analysis, testing, and formal verification on projects including adapting a codebase of hundreds of millions of lines for Clang/LLVM, fixing analyzer findings, and moving services from x86 to ARM. Its approach concentrates responsibility in a small expert team: agents learn proposed fixes from sampled examples, automated reviewers assess them, and experts approve reusable guidance rather than inspect every edit. Feedback and grouped reviews help direct human attention to ambiguous or novel cases.
For C and C++ modernization, the aim is to translate legacy software into Rust without changing its behavior. The process has progressed from producing code that would not compile to automating much larger migrations; the remaining challenge is proving the translations are sound. Teams organize work around user-visible features, validate them with scenario-based and differential tests, and increasingly use specifications and tools such as Verus to prove properties. More broadly, agents can infer specifications that guide testing and future changes, while engineers approve the properties those specifications should capture. This raises unresolved questions: whether AI-derived specifications reflect intended behavior, how to verify security and reliability as well as functionality, whether engineers will still need to read code, and how responsibility for code changes when agents can modify entire systems.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
New records show Medicare’s WISeR AI prior-authorization program has caused care delays, denials, operational problems, and reported patient harm. The EFF says the findings highlight the need for stronger legal limits on government use of automated decisions. (eff.org)
Anthropic has partnered with six biotech organizations and launched its own drug discovery programs to test Claude across the drug development process, from research and molecule design to clinical trials. The work has identified model improvements and could contribute to treatments for food allergies, childhood diseases, and cancers. (anthropic.com)
The supplied page contains no article text, so it provides no details to support the headline’s claim that China’s AI bubble has burst. The reasons and evidence behind that claim are unavailable. (finance.yahoo.com)
Survival LLM is a free, open-source AI assistant and emergency reference library that runs entirely on your computer, even without internet. It answers questions using sources such as FEMA and CDC guidance, and lets users inspect the cited passages. (brangus.io)
President Donald Trump says he will ban the term “artificial intelligence.” He wants the technology called “synthetic intelligence” instead. (thehill.com)
Donald Trump said he plans to ban “artificial intelligence” in federal documents and promote “super intelligence” instead. He has already directed the federal government to use the replacement term. (independent.co.uk)
Today's best Hacker News stories, summarized and screenshotted, one email a day.