AI Drug Reverses Aging Markers in Trial; OpenAI Warns on Pace

Insilico’s AI-Designed Drug Shows Biological Age Reversal in Early Trial

Insilico Medicine reported that its AI-designed drug candidate rentosertib showed signs of reversing markers of biological aging in an early trial. Originally developed for idiopathic pulmonary fibrosis (IPF), the drug was tested in a 42-patient trial last year that showed improved lung function. A new analysis of blood samples from that trial, published in Nature Biotechnology, used six independently built aging clocks — AI models that predict biological age from blood proteins — built by teams at Harvard, Oxford, Beijing, and Insilico itself. All six predicted a lower biological age for treated patients than for those on placebo, with the strongest effect showing a drop of three to four years by week 4, and one clock showing as much as six years.

AI-designed drug appears to turn back the body’s biological clock in early trial →

OpenAI Reports Research Agent Milestone and Warns on AI Pace

OpenAI reported that it has reached its goal of an automated research intern, a system that handles clearly scoped research tasks under human guidance, including tasks that would take an experienced researcher several days. The company did not share detailed validation, stating only that the milestone was met according to its measurements. Usage data showed that the median researcher at OpenAI spends more than $600 a day in inference at API prices, and the 90th percentile spends above $7,000. Since June, agent runtime has topped human working hours; as of mid-August, the research organization runs 3.1 agent workdays for every human workday.

OpenAI reports AI “research interns” and warns about its own pace at the same time →

Alibaba’s Qwen-Drive 1.0 Adds Spatial Awareness to Driving AI

Qwen-Drive 1.0 is a driving model developed by Alibaba’s research division that handles three tasks in one AI model: spatial perception of the environment, answering questions about traffic, and route planning. The researchers state that a text-image model does not automatically understand three-dimensional space just because it can describe pictures. The model builds on Qwen3.5-4B and adds two extra components: one generates a bird’s-eye-view map of the surroundings, and the other, the Planning Expert, uses internal model data to plan the car’s future movement. The team found that only when the vision-language model itself was trained on spatial tasks did performance improve significantly, meaning the ability to spatially understand traffic scenes has to be built in deliberately.

Qwen-Drive 1.0 tells you why it brakes, just don’t expect the explanation to match the maneuver →

Cathay Pacific and Google Cut Contrail Warming on Ultra-Long-Haul Flights

Cathay Pacific has flown more than 80 contrail-avoidance routes in a trial with Google that began in late 2025, adjusting cruise altitude to avoid thin bands of humid air where aircraft exhaust freezes into clouds that trap heat. Google’s satellite analysis estimates the trial produced roughly a 40% reduction in the warming impact of the contrails those flights would otherwise have made. The trial targeted ultra-long-haul services of more than 16 hours, the first time the technique has been tested at that duration. The Hong Kong to Singapore route, where persistent contrails form often, accounted for more than half the estimated reduction.

Cathay Pacific and Google partner on AI contrail avoidance for ultra-long-haul flights →

AI Coding Agents Push Companies to Build Software Internally

Nearly one-third (32 per cent) of respondents in a McKinsey research report said their organisations had decided against purchasing at least one software product or feature because they could build the functionality internally using agentic coding tools. The trend was particularly pronounced in technology and healthcare, followed by professional services and energy and materials. Forty percent of respondents at organisations with more than $1 billion in annual revenue said they were scaling AI agents in at least one function, up from 27 per cent a year earlier. AI high performers are twice as likely as other organisations to report scaling software coding agents and 2.7 times more likely to scale other agentic AI.

AI coding agents shift corporate tech spending: Companies build software internally over buying →