The period’s main signal is stricter proof for AI-built software. ConCovUp, RubricRefine, and MonitoringBench test agents against concrete failure modes: missed concurrent interactions, wrong tool contracts, and hidden…
Agent software work is moving toward checks tied to specific failure modes: live tool calls that return plausible wrong answers, monitors tested with narrow attack sets, and C/C++ libraries whose sequential tests miss…
The day’s strongest work treats AI coding as a governed engineering process. AutoMat tests scientific reproducibility, SAGA measures full agent latency, and RECAP records real prompt-to-edit traces.
AI coding agents now have enough local evidence to justify three concrete changes: recording prompt-to-edit traces during real development, testing scientific agents with claim-level reproduction tasks, and measuring…