Omnissa combines Horizon virtual desktops with NVIDIA vGPU to give AI agents controlled access to screen-based enterprise systems
The AMW Read
Omnissa extends VDI/GPU-virtualization infrastructure to support GUI-operating AI agents on API-less enterprise systems, an incremental but broadly applicable infra update tying agent workloads to dynamic GPU allocation.
Omnissa combines Horizon virtual desktops with NVIDIA vGPU to give AI agents controlled access to screen-based enterprise systems
Omnissa, the digital workplace company behind the Horizon virtual desktop platform, detailed four enterprise use cases pairing Horizon with NVIDIA's vGPU technology, which partitions a single physical GPU across multiple virtual machines. The combination lets IT teams provision and reclaim GPU-backed virtual workspaces on demand: AI developers spin up GPU-accelerated environments for model fine-tuning (a pharmaceutical example cut setup time from weeks to under an hour), while vision models reading GPU-rendered screens let AI agents operate legacy, API-less software such as ERP systems, core banking platforms, and industrial SCADA dashboards by parsing on-screen buttons, menus, and data states.
The approach targets a real gap in current agent tooling: most enterprise automation assumes API access, but a large share of business-critical software has none. By extending existing virtual-desktop session recording, access control, and audit policies to AI-agent sessions, Omnissa positions screen-operating agents as an extension of IT-managed compute rather than a new unmanaged surface, and it preserves live-session oversight so a human — a grid operator, a lawyer reviewing an AI-drafted merger agreement — can intervene mid-task without restarting the agent's work. It also reframes GPU virtualization as an agent-workload scheduling problem: allocate GPU performance to whichever user or agent needs it, then reclaim it once the task ends.
For builders, computer-use agents don't need a dedicated environment stack if an enterprise already runs virtual desktop infrastructure — governance, logging, and GPU provisioning can be inherited rather than rebuilt. For infrastructure buyers, GPU allocation policy, not just model choice, will decide which systems agents can touch: Omnissa's construction example, where a vision model checked GPU-rendered building information model screens for design conflicts, cut pre-construction review of hundreds of models from days to hours — a template for imaging, compliance, and other visually-driven enterprise workflows.
#AIAgents #EnterpriseAI #NVIDIA #GPUVirtualization #VirtualDesktopInfrastructure #ComputerUseAgents
