AWS Releases Pizza Bot, an Open-Source Background Task Manager for AI Agents

Developers at AWS have made public Pizza Bot, an open-source tool for running AI agents on background tasks and collecting their results through an inbox-like interface. The application lets agents execute scheduled or webhook-driven work, hand off subtasks to specialized agents, and wait for human sign-off before taking critical actions.

How Pizza Bot Works#

Pizza Bot functions as an email-style inbox for managing long-running agent workflows. Built with an Apache 2.0 license and a client-server design, it lets teams delegate sophisticated tasks without needing to monitor progress continuously. The interface organizes work into separate queues—All, Unread, and Action—while a separate Activity panel tracks tasks sent to specialist agents and their current status.

The tool supports multiple AI model providers, MCP servers, and custom Agent Skills. All agent state, message threads, checkpoints, files, and activity logs stay stored locally in a folder you control. Data only leaves your machine when sent to your chosen model provider or to explicitly enabled MCP servers, with passwords stored securely in your operating system's credential vault.

Design Philosophy#

The name echoes Amazon's two-pizza teams concept. AWS technologists Joseph Dolivo and Igor Fil explain the inbox metaphor as central to the architecture: "You don't send an email and then sit watching the outbox until the reply lands. Pizza Bot is shaped like an email client for the same reason: a thread is a unit of work you come back to rather than a session you have to attend."

Four core principles shaped the design—agents can consult multiple information sources, operate without direct human supervision, request approval before risky decisions, and alert users when tasks finish. The local-first approach means users retain sovereignty over what data leaves their system.

Origins and Reception#

The project began in April 2025 as a side effort called "JoeBot" to automate repetitive CRM data entry. As word spread, both technical and non-technical colleagues wanted to use it, prompting a broader rethink of the tool's design and capabilities.

Early responses have been largely enthusiastic, particularly around the asynchronous inbox model and local-first storage. Some observers on Hacker News praised the work while raising questions about the growing number of similar agent-management tools in the market, and whether an open-source project can build the same depth of integrations and community skills that the original internal version achieved.