A startup says top AI tools pass only about half of its cloud operations tests, and it plans to sell agents trained to do better in early 2027.
Vera Platforms, a startup founded by University of Michigan researchers, says the best-known AI tools get only about half of its cloud operations tests right. Its CEO, Ang Chen, says Claude and OpenAI's Codex pass roughly 50 percent. The company plans to release AI agents trained for this work in the first quarter of 2027.
DevOps is the work of putting software online and keeping it running. For many companies that means managing servers and storage rented from a cloud provider such as Amazon. Chen says live systems are unforgiving. In his words, a wrong change can take down a service, open a security hole or erase data, and some mistakes cannot be undone.
One example involved cleaning up old resources in Amazon's cloud. Amazon runs its services in many regions, and the default is called us-east-1. Chen says Claude and Codex did not look beyond that default. They removed only part of what needed to go, because they did not realize resources could sit in several regions.
His explanation is that these models have seen a lot of public code but far less of how real systems are operated day to day.
The test, called CloudOpsBench, includes tasks like deleting retired storage buckets and restoring automatic scaling on a database. Chen says senior infrastructure operators wrote the tasks. He also says the simulator mimics Amazon closely enough that an agent behaves the same there as on the real cloud.
The numbers and the simulator claim come from the company's own CEO. The material gives no independent results, no pricing and no details on how well vera-agents will score. The 2027 date is a target, not a finished product.
Anyone hoping AI can run a company's cloud should treat it as a helper that needs supervision. Try it first on a safe copy of your setup, not on the live one, and have a person check its changes before they go out.
Before letting AI change your company's cloud setup, test it on a safe copy first and have a person review the changes.
Otto, a new customer-tracking app for field sales teams, lets reps speak their notes instead of typing them. Price and launch details are not yet known.
Read more: Otto launches a sales app you talk to instead of typingA former Nvidia product manager has started Udu Technologies to let African startups and institutions rent AI computing power instead of buying it.
Read more: Ex-Nvidia engineer starts company to bring AI computing power to AfricaEA has re-released 25 classic PC games at a discount until November 19, and Bethesda has finally given The Elder Scrolls VI a subtitle: Sentinel.
Read more: EA brings 25 classic PC games back, and Elder Scrolls VI gets a name
