Cognition has modernized macOS app development for AI agents by running each Devin session inside a fresh macOS VM hosted on Amazon EC2 Mac bare metal instances. This architecture eliminates the need for managing physical Mac hardware while enabling rapid, isolated, and fully functional macOS environments to build, run, and test iOS and macOS applications.

  • Fast startup of isolated macOS VMs for each AI coding session
  • Runs on Apple silicon-backed bare metal EC2 Mac Dedicated Hosts
  • Scales globally with automated host allocation and VM snapshot management

Infrastructure signal

Cognition deploys macOS workloads on Amazon EC2 Mac instances running on bare metal Apple hardware, combined with the AWS Nitro System to deliver dedicated hosts. Each macOS host supports two parallel macOS guest VMs, adhering to Apple's license allowance. Infrastructure leverages Dedicated Hosts with a 24-hour minimum lease to balance cost against session lifecycles. Automated Auto Scaling handles host provisioning in regions without fixed capacity.

The VM lifecycle is separated from host allocation to optimize resource usage. Each session boots a prepared macOS VM image using Apple's Virtualization framework, while disk I/O is managed through Network Block Device (NBD) snapshots that facilitate rapid snapshot restore and persistence. Multi-region deployment enhances global reach and resilience while leveraging AWS’s scalable backend for cloud-native macOS development environments.

Developer impact

By granting Devin, the AI software engineer, a dedicated macOS VM per session, Cognition enables full macOS/iOS app build, run, and test capabilities without relying on customers’ physical Macs or CI pipelines. Developers and AI agents gain access to Xcode tools and iOS Simulators inside these ephemeral environments, allowing for comprehensive automated behavioral testing including accessibility-driven UI interactions.

The rapid VM startup time—under a minute—drastically improves developer workflows and continuous integration speeds compared to traditional Mac hardware setup. This instant provisioning helps scale macOS app development within AI workflows, minimizing friction and infrastructure overhead associated with dedicated Apple hardware management.

What teams should watch

Teams planning to integrate AI-driven development agents for macOS and iOS should closely evaluate cloud-based Mac solutions like EC2 Mac bare metal instances for scalability and reliability. Licensing limits—two virtual macOS guests per host—and 24-hour Dedicated Host minimums mandate careful session orchestration to optimize utilization and cost.

Observability around VM snapshot management, networking gateways, and accessibility-based controls is key to maintaining reliable test automation and user session interactivity. Infrastructure teams need to design automated scaling and host lifecycle management strategies that decouple session duration from host leasing constraints to avoid overprovisioning costs.

Source assisted: This briefing began from a discovered source item from AWS Architecture Blog. Open the original source.
How SignalDesk reports: feeds and outside sources are used for discovery. Public briefings are edited to add context, buyer relevance and attribution before they are published. Read the standards

Related briefings