Ari Weinstein explains why Computer Use has changed dramatically through better models, self-debugging, richer UI representations, generated code, and improved harnesses. The conversation then turns to trust, consent, software testing, and the path from human-level speed to superhuman operation. Nikunj Handa details OpenAI’s API work on asynchronous tool calls, WebSockets, low-latency inference, the rapidly built Decisions API, prompt caching, compaction, and higher-level agent abstractions. Together, the guests describe OpenAI’s attempt to make agents reliable, fast, persistent, and easier to build on.