Recommend a hotel from client notes using Codex
OpenAI Codex just landed on Mac and PC, and travel advisors should pay attention. Task: A travel advisor used OpenAI Codex Desktop to analyze a local folder containing client notes (e.g., Word docs) and cross-reference them with live web data from hotel brand sites. The agent identified client preferences from past trip fragments, checked current availability and new properties, and generated a comparison of three matching options. The human specified the folder and brand but did not provide destination details, relying on the agent to infer context from files.; tools: Codex. Human role: The original post does not say how a person stayed involved.; limitations: Codex Desktop requires a ChatGPT subscription (free tier has limited usage; Plus is $20/month). The app is new and may have rough edges. Human review is mandatory before sending recommendations to clients, as the output may contain inaccuracies or require pricing adjustments based on exact traveler counts. Setup takes a few minutes..
Tools
Published on Agent Work Atlas:
Overview
OpenAI Codex just landed on Mac and PC, and travel advisors should pay attention.
Workflow
A travel advisor used OpenAI Codex Desktop to analyze a local folder containing client notes (e.g., Word docs) and cross-reference them with live web data from hotel brand sites. The agent identified client preferences from past trip fragments, checked current availability and new properties, and generated a comparison of three matching options. The human specified the folder and brand but did not provide destination details, relying on the agent to infer context from files.
Reported result
The agent produced a ranked list of three hotel options with availability signals and pricing estimates for a specific date window, along with a detailed Word document. The author notes the output is not perfect and cannot be sent directly to clients without review, but it saves time and points in the right direction. No measured efficiency metrics or final booking outcomes are reported.
Human role
The original post does not say how a person stayed involved.
Limitations
Codex Desktop requires a ChatGPT subscription (free tier has limited usage; Plus is $20/month). The app is new and may have rough edges. Human review is mandatory before sending recommendations to clients, as the output may contain inaccuracies or require pricing adjustments based on exact traveler counts. Setup takes a few minutes.
Try it
Act as a travel advisor using OpenAI Codex Desktop. I have a folder at [replaceable folder path] containing client notes in [replaceable file format, e.g., .docx]. The client is interested in [replaceable brand, e.g., Belmond] but I haven't specified a destination. Please: 1. Read all files in the folder to identify the client's travel preferences, past trip patterns, and budget constraints. 2. Browse the web to check current availability and new properties for [replaceable brand] that match these preferences. 3. Create a comparison table of three top options, including destination, why it matches, availability signals, and potential watchouts. 4. Generate a detailed Word document summarizing the recommendation. Do not send this to the client. First, perform a dry run on one sample file to verify you can extract preferences correctly. Show me the extracted preferences and the first draft of the comparison table for my approval before finalizing the Word doc. If you cannot access the folder or browse the web, stop and ask for clarification.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.