How we built a platform in 4 weeks
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
An Atlassian engineer shares how their Zero Trust Compute team built a Firecracker-based microVM orchestration platform called Fireworks in just four weeks using Rovo Dev, Atlassian's AI coding agent. The workflow involves running multiple agents in parallel across branches, giving agents full SDLC access including deploying to isolated dev shards, writing and running e2e tests, raising PRs, and reading CI pipeline output. Key practices include AI-written e2e tests as the primary validation harness, black-box output validation instead of code review, adversarial subagent PR reviews, progressive canary rollouts, and investing in repo-specific skills and agent configurations. The author argues that developer roles shift toward architecture and design, while agents handle implementation details, and that teams must also embrace AI-assisted reviews to avoid human bottlenecks.
Table of contents
What We’re BuildingThe Setup Is SimpleTreat It Like a Real EngineerContext Feedback LoopsHow We ValidateYour Team Needs to Be Agentic TooInvest in Your AI SetupYour Role ChangesTalk to your code through the AgentMitigate risk with strong design, not manual reviewWhat’s working really well with Rovo DevSort: