Atlassian Builds AI Agent Execution Platform Fireworks in Just Four Weeks

Zero to One in Four Weeks: Atlassian's Rapid AI Platform Development

Atlassian recently revealed a remarkable engineering feat on its official blog Work Life — a team built a secure execution platform called "Fireworks" from scratch in just four weeks. The platform serves as the core engine of Atlassian's AI agent infrastructure, powering the backend operations of its AI product Rovo. What makes this even more noteworthy is that the development process itself heavily relied on Rovo's AI capabilities, creating what the team calls "Rovo Dev Driven Development" — an entirely new development paradigm.

Fireworks: Where MicroVMs Meet Kubernetes

From a technical architecture standpoint, Fireworks addresses a critical infrastructure challenge — how to securely and efficiently run a Firecracker microVM (microVM) orchestrator on Kubernetes clusters.

Firecracker is an open-source virtualization technology from AWS, renowned for its ultra-fast startup times and minimal resource footprint. Originally developed for serverless products like AWS Lambda and Fargate, Atlassian chose Firecracker as the execution environment for AI agents, clearly valuing its dual advantages in security isolation and performance.

The Fireworks workflow is clean and efficient: users simply submit an OCI container image and an execution command, and the system automatically spins up a Firecracker microVM to run the task. This design combines the convenience of containers with VM-level security isolation, providing a flexible yet secure sandbox environment for AI agent code execution.

This architectural choice carries profound implications. AI agents often need to run user-specified code or invoke external tools during task execution, and traditional container isolation may pose security risks when facing malicious code or unexpected behavior. The hardware-level isolation provided by microVMs significantly raises the security boundary — a critical consideration for enterprise-grade AI agent platforms.

"Rovo-Driven Development": AI Accelerating AI Infrastructure

The most compelling concept in this development effort is Atlassian's "Rovo Dev Driven Development" model. In essence, the team extensively used its own AI assistant Rovo to support the development work while building the AI agent execution platform.

This approach of "using AI to build AI infrastructure" not only validated Rovo's practical value in real-world engineering scenarios but also made the highly ambitious four-week delivery timeline possible. From code writing and architecture design to troubleshooting, the AI assistant served as an accelerator throughout the entire development lifecycle.

This practice also echoes a major trend in the software engineering field: AI coding assistants are evolving from "supplementary tools" to "development drivers." When teams deeply embed AI into their development workflows, productivity gains are no longer linear — they can deliver transformative leaps.

The Enterprise AI Agent Race Intensifies

Atlassian's move also reflects the increasingly fierce competition in the enterprise AI agent space. Rovo, launched by Atlassian in 2024, aims to provide enterprise users with intelligent search, knowledge management, and automated agent capabilities. The launch of the Fireworks platform means Rovo's agent capabilities will gain more powerful and secure underlying support.

Currently, Microsoft Copilot, Google Agentspace, ServiceNow, and numerous startups are all actively positioning themselves in the enterprise AI agent arena. The competitive focus among vendors is gradually shifting from "model capabilities" to "execution infrastructure" — whoever can provide AI agents with a more secure, reliable, and efficient runtime environment will hold the advantage in this race.

Atlassian's decision to build its execution engine on Firecracker microVMs sets a high bar for security, which is particularly crucial for serving enterprise customers. After all, one of the biggest concerns enterprise users have about AI agents is security and controllability.

Looking Ahead: A Paradigm Shift in AI Engineering

Building a production-grade platform in four weeks is itself a powerful testament to AI-empowered software engineering. Atlassian's experience demonstrates that when AI tools are deeply integrated with engineering teams, traditional project timelines and resource estimation models may need to be redefined.

Looking forward, "AI-driven development" is poised to become the standard model for more technical teams. At the AI agent infrastructure level, secure execution environments, microVM orchestration, and the convergence of containerization and virtualization will continue to be key areas of industry focus. Whether Atlassian can stand out in the enterprise AI agent race with its Fireworks platform remains worth watching.