A formal test found older language-model agents solved fewer tasks when acting than when writing complete plans. The failures expose what an agent evaluation must check, while leaving today's reasoning models an open question.
Javo reports higher sales in a live trial, planning agents struggle with legal actions, and IBM improves reliability through learned memory. OpenAI discloses monitoring failures; DeepSeek opens DeepInfer.
Cloudflare's AI code reviewer grew more expensive as its models improved. The company cut average cost by 68 percent and median latency by almost two thirds, then had to prove that the cheaper reviewer had not simply stopped finding problems.
An AI voice agent gave human sellers better leads but they closed fewer of them, while a robotics index placed the greatest physical-AI exposure in lower-paid work. A pre-registered audit found that three common measures of model confidence disagree.
Nine frontier agents produced 135 follow-up papers for unpublished research directions from national-lab scientists. None was honest, complete and scientifically sufficient at the same time.
A realistic delivery benchmark finds the best AI agent resolves only 14 of 95 customer projects, while nine frontier agents fail every scientist-set follow-up task. Microsoft and CMU respond to a widening experience gap in AI-assisted software engineering.
A ten-agent experiment finds that one persistent adversary can make benign agents privately adopt a false belief. Public answers reveal less of the collapse than internal probes do.
Anthropic's economists report that junior technical hires fell from 27% to 9% of hires while agent hours reached six times human hours. The report is one company's account of itself, but it shows what happens when human attention becomes the bottleneck.
Ant Group opens a 560-billion-parameter hybrid reasoning model, a new benchmark finds higher-order theory-of-mind shortcuts, and offline speech reaches a $20 ESP32.
Language models looked capable of following beliefs inside beliefs. HopToM shows that many benchmark questions let them answer by taking a shallower route, then offers a cheaper way to reason through the full chain.
Liquid AI released an open 24-billion-parameter mixture-of-experts model for on-device agents. New research found compact structure in millions of human choices and a causal spatial workspace inside multimodal models.
Meta released Muse Wonder for long-running personal and professional agents, while Anthropic tested whether small models could replicate across cheap IoT hardware. Two preprints show how task context changes attention-head roles and how a fresh user turn can redirect failed reasoning.