Book a New Delhi to Mumbai flight with computer use
The author ran Claude's computer use in a Docker demo to book a flight from New Delhi to Mumbai; Claude opened Firefox and started the search, but the outcome isn't shown. Task: Suggested implementation: Run Anthropic's computer use demo in a Docker sandbox with an API key. Prompt Claude to book a flight from New Delhi to Mumbai for a specific date. Claude opens Firefox and navigates to the travel site. The human monitors the session and provides follow-up prompts to resolve UI obstacles like sign-in popups.; tools: Claude computer use. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include a Docker environment, Anthropic API key, and network access to travel sites. Constraints: The agent may stall on complex UI elements like popups, requiring human prompts to proceed. Anthropic advises using sandboxed containers and mandatory human sign-off for financial transactions. Source reliability is low as the original URL returned a 403 error..
Tools
Published on Agent Work Atlas:
Overview
The author ran Claude's computer use in a Docker demo to book a flight from New Delhi to Mumbai; Claude opened Firefox and started the search, but the outcome isn't shown.
Workflow
Suggested implementation: Run Anthropic's computer use demo in a Docker sandbox with an API key. Prompt Claude to book a flight from New Delhi to Mumbai for a specific date. Claude opens Firefox and navigates to the travel site. The human monitors the session and provides follow-up prompts to resolve UI obstacles like sign-in popups.
Reported result
Outcome unknown due to missing source access. Based on the provided case summary, Claude did not complete the booking but successfully filled the search form after human intervention to close a popup. No verified metrics or final success rates are available. Proposed acceptance check: Verify that the agent can navigate to the search page and fill date/city fields without crashing.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include a Docker environment, Anthropic API key, and network access to travel sites. Constraints: The agent may stall on complex UI elements like popups, requiring human prompts to proceed. Anthropic advises using sandboxed containers and mandatory human sign-off for financial transactions. Source reliability is low as the original URL returned a 403 error.
Try it
Act as a QA engineer testing an AI agent's web navigation capabilities. I want to run a dry test of Anthropic's computer use demo in a Docker sandbox. Context: We are evaluating if the agent can handle basic flight search forms on makemytrip.com. Inputs: [API_KEY], [DOCKER_IMAGE_NAME], [TARGET_URL: makemytrip.com], [SEARCH_PARAMS: New Delhi to Mumbai, Dec 1, 2024]. Steps: 1. Start the Docker container with the provided API key. 2. Launch the agent and instruct it to open Firefox and navigate to the target URL. 3. Ask the agent to fill in the origin, destination, and date fields. 4. Monitor for stalls (e.g., popups). If stalled, provide a specific prompt to close the popup. 5. Stop the agent before any payment or final booking confirmation. Output: A log file showing the agent's actions and a screenshot of the filled form. Success Metric: The form fields are correctly populated without manual browser control. Safeguards: Do not allow the agent to proceed to payment. Require my explicit confirmation before sending any prompts that might trigger a transaction. If the agent fails to navigate, report the error and stop. Ask me for the API key if it is not provided in the environment variables.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.