As an alternative to defining instructions manually for agents, you can use Arc. This is a quick way to use AI to generate an agent application for you.
With this approach you provide:
- A clear prompt describing what you want your agent application to do.
- Optional knowledge documents that provide technical details for the agent application, which can include technical details for external services to utilize.
Steps to generate agents
To generate agents:
- Using the agent builder, create an agent application.
- Click the Start with AI button in the root agent or click the Arc button at the top of the builder.
- Provide a clear goal for your agent application.
- Optionally add knowledge documents using the Upload button. These documents help teach your agent how to handle a variety of scenarios. You can upload files like tool catalog, requirements, transcripts, external services, and FAQs. These files can be used to provide specifications for tools and back end integrations (PRD, PDF, Markdown, OpenAPI). You do not need to attach any files; a clear natural language prompt works well on its own in many cases.
- Submit your prompt and upload files. Arc will start streaming its chain of thought. Once complete, agents are created with instructions, tools, and callbacks as necessary.
Example prompt for customer support and flight rebooking:
Create a customer service agent for an airline that can check flight status,
assist passengers with rebooking flights, and answer questions about baggage
policies.
Example prompt for customer order status
(uploaded files: return_policy.md, orders_api.json):
Create an e-commerce assistant based on the attached return policy
and OpenAPI catalog. The agent should answer customer shipping FAQs,
check order delivery status using the orders tool, and route return
requests to a live agent if the item was purchased more than 30 days ago.