[OpenAI Report: AI Programming Agents Can Accelerate Scientific Software but Cannot Assess Scientific Accuracy] Deep Tide TechFlow reports that on August 1, a field report by OpenAI and academic partners revealed that programming agents like Codex and Claude Code can accelerate outdated scientific software by over 60 times. RustQC reduced runtime from 15 hours and 34 minutes to 14 minutes and 54 seconds, while HelixForge was 59.6 times faster than BamSurgeon. However, the agents cannot reliably determine whether the results are scientifically accurate and often confidently produce erroneous code, requiring humans to define testing and validation standards.