Duckbill is built on the premise that large language models are capable of planning but incapable of independent real-world execution. While an LLM can identify that a flight should be refunded or that an insurance claim was denied, it cannot sit on hold for 45 minutes to speak with a customer service representative. Duckbill addresses this 'last mile' problem by combining an AI-driven interface with a workforce of background-checked human specialists.

Founded in 2022 and backed by investors including General Catalyst and Red Antler, the company positions itself as an executive assistant for personal life. The core product is a subscription service where users submit tasks via text, email, or a mobile interface. Once a request enters the system, AI assists in breaking the task into logical steps and conducting initial research. However, the final execution—the phone calls, the arguments with insurance providers, and the coordination of local services—is handled by over 200 vetted human operators.

The hybrid execution model

The platform uses a tiered approach to task management. AI handles the intake and planning, which reduces the cognitive load on human operators. This allows the human specialists to focus entirely on the parts of a task that require empathy, persistence, or navigation of non-digital bureaucracies. For example, if a user needs to find a doctor, the AI identifies in-network providers, while the human operator calls each office to confirm actual availability and book the slot.

This hybrid model creates a higher completion rate than autonomous agents. Most digital-only assistants struggle with the nuances of human interaction, such as a pharmacy claiming a prescription is ready when it has not yet been processed. Duckbill specialists handle these discrepancies by following up until the task is marked as finished. This is why their marketing emphasizes 'done, actually done' rather than just providing reminders or research summaries.

Integration with the agent stack

A significant development for Duckbill is its support for the Model Context Protocol (MCP). This allows users of AI models like Claude to delegate tasks directly from their chat interface to the Duckbill execution engine. In this context, Duckbill acts as a tool that an LLM can call when it reaches a real-world bottleneck. This bridge turns a purely conversational AI into an agent capable of affecting the physical world through human proxies.

For developers building their own AI agents, Duckbill offers a way to increase task success rates. Instead of an agent failing when it encounters a phone tree, it can hand the task off to Duckbill's human-in-the-loop system. This positioning moves Duckbill from a consumer concierge service toward a infrastructure layer for the broader agent ecosystem.

Market position and pricing

Duckbill operates on a subscription-based model with different bandwidth tiers. The Individual plan is priced at $99 per month for roughly four to six tasks, while Household and Household Plus plans offer more capacity. This pricing reflects the high cost of human labor compared to purely algorithmic solutions. The company targets individuals with high cognitive loads, such as startup founders or parents, who are willing to pay for the assurance that a task will be completed without their further involvement.