AI / Core OpenAI Codex Application Fundamentals Interview Questions
What is the OpenAI computer use capability and what does it enable?
Computer use is a capability that allows an OpenAI model to interact with a computer's graphical interface - clicking buttons, filling forms, navigating browsers, and reading screen content - just as a human user would. It is available as a built-in tool in the Responses API and as a core feature of the Codex CLI.
| Capability | Example use case |
|---|---|
| Browser navigation | Research competitor products, scrape structured data, fill web forms |
| GUI interaction | Test desktop applications, automate legacy software with no API |
| Screen reading | Extract data from PDFs opened in viewers, read application states |
| Keyboard and mouse control | Any repetitive UI workflow |
| Multi-application workflows | Copy data from one app to another |
# Computer use via the Responses API: response = client.responses.create( model="gpt-5.4", # gpt-5.4 has strong computer use support tools=[{ "type": "computer_use_preview", "display_width": 1280, "display_height": 800, "environment": "browser", # or "desktop", "linux", "windows" }], input="Go to github.com/openai/codex, find all open issues labelled bug, and return a summary.", truncation="auto", ) # In Codex CLI - computer use is built into the tool automatically: # codex # > Open the browser and check if our staging deployment is up at staging.example.com # Codex will: open browser, navigate to URL, read the page, report status # Safety considerations: # 1. Sandbox environments recommended # 2. Require approval for form submissions, purchases, account changes # 3. Log all actions taken by the computer use agent
Approval modes and computer use: in Codex CLI, the full-access approval mode enables computer use with network access. The auto mode (workspace-scoped) limits computer use to the local workspace. For production use, always implement human-in-the-loop approval for actions that modify state or involve external services.
Which OpenAI model has particularly strong computer use support as the default in Codex App and CLI?
What type of approval is recommended for computer use agents performing actions like form submissions or account changes?
More Related questions...
