Build Apps While You Sleep: Autonomous Engineer Demo with Apurva Desai

[Earl Duque]

You know how the backlog goes. A requirement lands, it sounds simple, and a few days later you're several ACLs deep wondering how a small form request turned into a mini project. The Brazil release brings something that takes a swing at that pile: the Autonomous Engineer.

You might know it by its internal name, Build Agent Spec Mode. It's an agentic engineer that slots into the SDLC you already run.

Key features and workflow

• From PRD to plan: Upload your business requirements (a PDF PRD works fine) or connect through MCP to your Agile system. The agent reads the specs, and where something is vague or missing, it interviews you until the requirements hold up.
• Instance-aware execution: Before it builds anything, it checks your instance context, so it knows which tables and configurations already exist and won't rebuild what's there. Then it lays out a sequenced dependency graph of work items so tasks can run asynchronously, in parallel.
• Automated testing and artifacts: It generates and runs ATF tests against every acceptance criteria. When the work is done, it rolls everything into a batch update set you can deploy.
• You stay the director: It won't invent requirements or push straight to production. Give it two ACL requirements that contradict each other and it parks the item in draft and asks you how to proceed.

Work while you sleep

Asynchronous execution is where this changes your day. Review the generated plan, click Execute, and step away. Let it run overnight and you come back to a dashboard of completed stories, ATF results, and the decision points it needs you to weigh in on. Less time grinding out manual configuration, more time on the architecture calls that actually need a human.

Availability and models

Autonomous Engineer is launching through the new AI Workflow Factory for the Prime SKU. A freemium tier for Personal Developer Instances is coming, along the same lines as Build Agent. You can also bring your own model and configure the LLMs you prefer through the AI Admin Hub.

Want to learn more?

• Watch the full Developer Passport episode (embedded at the top of this post) and catch Apurva Desai demo the Autonomous Engineer building a Lego tracking app live.
• Grab your complimentary pass for now[dev]26 this November.
• Subscribe to our YouTube channel for the rest of our Brazil release deep dives.

So what would you point it at first? I'm curious whether folks reach for net-new builds or finally clear out the backlog of small stuff nobody wants to pick up.