DAIR.AI describes ActiveSaddler, a paper from Microsoft and colleagues that changes which scenarios are used to optimize an agent harness. With the same optimizer, test Pass@1 improves by 4.4 points on GAIA2 and 7.5 points on Terminal-Bench 2.0 versus a fixed scenario order.
In an interview with Alex Heath, Satya Nadella said Microsoft Autopilot’s AI chief of staff should also serve consumers’ personal lives, citing more than 100 million consumer subscribers. The agent is described as built on a hardened version of OpenClaw, with its own computer, workspace, and memory.
Omar Shahine said a preview of Microsoft Autopilot, a persistent personal agent, is reaching Microsoft’s first customers and is built on OpenClaw. Peter Steinberger said work with Microsoft to prepare the codebase for large-scale deployments has been underway since March.
CASD has an off-the-shelf coding agent analyze a full set of agent logs and write the findings as rules in one prompt, with no environment access or validation data. DAIR.AI says one pass beat GEPA and SkillOpt on the reported benchmarks at about $1.60 per optimized prompt.
A paper from Microsoft and colleagues reports that the best model answers 59.7% of Taste-Bench questions correctly when choosing the better open direction in long engineering and research tasks. Forks decided by later evidence are much harder, a larger reasoning budget does not raise accuracy, and distilling a teacher’s outcome judgment improves held-out SWE-bench Pro success.