Take orders for a sushi chain with an agent
The poster built a Claude agent that takes orders for a seven-location sushi chain, using menu, allergen, delivery zone, hours and promo data. Task: Suggested implementation: Integrate Instagram DMs via Meta API with Claude Sonnet 4.6. Configure a knowledge base containing menu items, allergens, delivery zones, hours, and conversation rules for multiple locations. The agent should handle chat interactions, collect order details, and generate payment links. Human staff must monitor for non-text inputs or complex exceptions.; tools: Claude. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include stable Meta API access and current menu/allergen data. Operating costs depend heavily on prompt caching efficiency to avoid reprocessing fees. Integration uncertainty exists regarding payment link generation and non-text message handling. Human approval is required for all financial transactions and exception cases..
Tools
Published on Agent Work Atlas:
Overview
The poster built a Claude agent that takes orders for a seven-location sushi chain, using menu, allergen, delivery zone, hours and promo data.
Workflow
Suggested implementation: Integrate Instagram DMs via Meta API with Claude Sonnet 4.6. Configure a knowledge base containing menu items, allergens, delivery zones, hours, and conversation rules for multiple locations. The agent should handle chat interactions, collect order details, and generate payment links. Human staff must monitor for non-text inputs or complex exceptions.
Reported result
Actual outcomes are unknown as the source is inaccessible. Proposed acceptance check: Measure order accuracy rate and human takeover frequency over a 50-conversation pilot. Target: >95% accuracy for plain-text orders with minimal manual intervention.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include stable Meta API access and current menu/allergen data. Operating costs depend heavily on prompt caching efficiency to avoid reprocessing fees. Integration uncertainty exists regarding payment link generation and non-text message handling. Human approval is required for all financial transactions and exception cases.
Try it
Act as a technical lead for a sushi chain. I need to pilot an AI order agent for one location. Inputs: - [replaceable: Instagram Business Account ID] - [replaceable: Menu CSV with allergens, prices, and prep times] - [replaceable: Delivery Zone JSON] - [replaceable: Bank Payment API Key] Steps: 1. Configure Claude Sonnet 4.6 with the provided menu and rules as a system prompt. 2. Set up a webhook to route Instagram DMs to the agent. 3. Implement logic to extract address, quantity, and items from text. 4. Generate a payment link upon order confirmation. 5. Flag non-text messages or unclear inputs for human review. Deliverables: - A working webhook endpoint. - A log of 10 simulated conversations. - A summary of token usage and estimated cost. Quality Check: - Run a dry run with 5 test messages. - Verify order accuracy and payment link validity. - Require my confirmation before connecting to live Instagram traffic. If inputs are missing, ask for the specific file format or API documentation needed.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.