OpenAI Coding Agents Accelerate Research Workflows
OpenAI is embedding coding agents across its artificial-intelligence research workflow, from building research infrastructure and running evaluations to analyzing experiments and troubleshooting internal systems. The shift matters because agents are moving beyond code assistance to complete well-defined assignments that could take a skilled researcher several days. OpenAI says humans still set research priorities, judge results and decide whether models should be scaled, paused or deployed, underscoring that faster execution has not removed the need for human oversight.
In data released on September 6, 2026, OpenAI said the median researcher was using more than $600 a day of agent inference at API prices by mid-August, while the 90th-percentile user exceeded $7,000. The research organization logged 3.1 agent-workdays for every human workday, based on an eight-hour day. Experiments per active experimenter reached their highest level in August since tracking began in January 2025. Success rates improved from January through July, although more than half of successful four-to-eight-hour tasks over the past six months still required at least one human intervention.
All Coverage
2 original reportsThe Backstory
The history behind this eventOpenAI Says Enterprises Shift AI From Assistance to Execution
Companies are moving generative AI beyond one-off writing and search tasks toward agentic systems that can complete multistep work across business processes. OpenAI’s report, released December 8, 2025, combines deidentified customer usage with a survey of 9,000 workers at nearly 100 enterprises. It tracks how ChatGPT and Codex are becoming embedded in repeatable workflows, making organizational readiness — including governance, training and system integration — a growing source of competitive advantage as model access becomes more widespread.
Weekly messages in ChatGPT Enterprise rose about eightfold in the year since November 2024, while the average worker sent 30% more messages. Weekly users of Custom GPTs and Projects climbed 19-fold in 2025, and those tools handled about 20% of Enterprise messages. Codex engagement accelerated over the six weeks covered by the report, with weekly active users doubling and messages rising roughly 50%. Firms in the top 5% of adoption intensity generated twice as many messages per seat as the median enterprise and seven times as many messages to GPTs.
OpenAI Finds Coding Agents Speed Scientific Software Development
Scientific computing underpins research across academia and industry, but many widely used tools began as code for a paper, built by small teams with limited time for packaging, testing, optimization or maintenance. The problem is acute in genomics, where sequencing costs have fallen faster than downstream analysis costs, increasing the burden of storage, compute and specialized labor. Coding agents matter because they could reduce that engineering bottleneck while allowing scientists to focus on validation and discovery.
OpenAI released the exploratory report on July 28, 2026, covering eight projects focused mainly on life sciences. Five used Codex alone and three combined Codex with Claude Code. In one case, RustQC processed a dataset of 186 million sequencing reads, cutting summed sequential runtime to 14 minutes and 54 seconds from 15 hours and 34 minutes, a more than 60-fold improvement, while reducing disk traffic to 0.1 terabytes from 2.5 terabytes. OpenAI said the results were contributor-reported and case-specific, with human validation still essential.
Subscribe to Mark Radar Weekly
Every Friday, the week's strongest signals in your inbox. Unsubscribe anytime.
If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →