Why OpenAI Is Racing to Make Computer Use Superhuman
OpenAI's Ari and James argue computer use is approaching superhuman speed, works universally without integrations, and now needs enterprise controls, better UX and proactivity to reach mainstream consumers and businesses.
Watch on YouTube ↗Charlie's summary
- ▶ 8:05
Superhuman computer use is a coming sea change: once the agent is faster than experts, doing things yourself becomes optional
Ari says computer use is already faster than the average human and is “getting really close” to being faster than an expert user. When that happens, anything you could do manually, the agent can do faster, so by default you would hand it off. Speed matters because nobody wants the feeling of “I could have just done that myself.” Ari thinks this changes how people interact with computers.
- ▶ 10:06
Enterprise adoption hinges on dials for human-in-the-loop control and auditability, not just model alignment
James says that as agents outrun human observation, businesses need to set how much human oversight there is. OpenAI's enterprise controls include deterministic sandboxes (allowed apps and sites), approval requirements for actions like downloads, uploads or new websites, and an auto-approve model, Guardian, that checks every action. Consumers may want no prompts, while enterprises choose which actions go through.
- ▶ 15:44
Computer use is universal: it needs no connector, because it uses software the way a human does
Ari points out that, unlike other tools, computer use requires no developer work such as building an MCP server. It works with every application by operating it as a person would, so the agent gets all the capabilities you have. James adds that he no longer separates computer-use tasks from other tasks. It is the model's universal API, and complex tasks will touch it at some point.
- ▶ 20:50
Start with boring system-of-record workflows: procurement, finance reconciliation and sales prospecting
James sees the first enterprise wins in tedious work across ERPs, CRMs and other systems of record, many without APIs. Examples: replacing the giant procurement spreadsheet handed from team to team, reconciling ledgers at month-end, and prospecting into a CRM. Creative work like circuit design, architecture and game design comes next, as companies learn the model's capabilities.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.