Why Dwarkesh is Wrong about Computer Use + How OpenAI shipped its Jev competitor in 1 Week

In recent months, the community that follows advances in AI’s use of computers has seen a marked acceleration in progress, a sharp contrast to the earlier perception that the field was moving slowly. Ari Weinstein, co‑founder of Sky and lead on the recent “Computer Use” developments, explained that integrating screenshots, accessibility data, the DOM, and tools such as Playwright has enabled agents to learn how to debug and recover from errors more quickly. Moreover, the combination of generated code with these visual resources is dramatically cutting the time required for an agent to complete complex tasks, approaching a “superhuman” level of software manipulation.

Meanwhile, OpenAI unveiled a very fast version of its Decisions API, part of a new developer stack that includes asynchronous tool calls, mid‑interaction flow control, WebSockets, ultra‑fast inference, and prompt caching. According to Nikunj Handa of the API team, the implementation was completed in a week, demonstrating the company’s ability to replicate the Jev platform’s features with a wrapper built on its Luna model.

This news is significant because it shows AI moving from an experimental tool to a practical, reliable assistant for everyday users, potentially transforming how we interact with software and accelerating the adoption of autonomous agents in production environments.

Read the original article on Latent Space

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.