Skip to main content

Agent Tasks

Instead of breaking a task into several separate prompts, you can describe a complex request in one message with multiple parts, and the AI works through each part in its response.

Writing a Multi-Part Request​

Combine everything you need into a single, clear prompt. For example:

Plan my business trip: GVA to Tokyo next week, return 2 days later.
Need a hotel in Shibuya and a good Japanese restaurant for lunch when I arrive.

The AI addresses each part of the request — flights, hotel, and restaurant — in a structured response, rather than only answering the first thing you asked.

What this doesn't do

The AI puts together a plan based on your request; it doesn't book flights, reserve hotels, or place orders on your behalf. Treat the output as a draft plan to review and act on yourself.

Getting Better Results​

  • List every requirement up front. If you leave out a constraint (budget, dates, location), add it to your next message rather than starting over.
  • Ask for a specific structure. "Give me this as a numbered list with one section per topic" produces an easier-to-follow answer than an open-ended request.
  • Switch to a more capable model for harder requests. For requests with many moving parts, the Thinking model (see Key Features → Model Switching) works through the reasoning more thoroughly than the Standard model.

Refining the Result​

Treat the first response as a draft. Follow up in the same chat to adjust one part without losing the rest:

You: [multi-part request above]
AI: [structured plan covering flights, hotel, restaurant]
You: "Change the hotel to something closer to Shinjuku instead."
AI: [updates just the hotel portion, keeps the rest]

This uses the same multi-turn behavior described in Key Features → Multi-turn Conversations.

When to Use It​

Agent Tasks is useful for planning, itineraries, and any request that naturally breaks into several related sub-questions — when you'd rather get one organized answer than run several prompts yourself.