Stop managing infrastructure: A new way to deploy AI agents and models
Aug 18, 2026Abdelrahman (Abdo) Mahmoud, Cathie Bozhenko
Summary
Introduces the Workload API, a new system for deploying and managing AI services. Key technical contributions include the Artifact, Workload, and Protons object model for describing and running containerized services, a CLI and REST API for deployment and management, integration with DataRobot's credential store for secrets, support for NVIDIA NIM and Hugging Face model deployments with efficient weight caching, and unified observability and governance features for AI workloads.
More from the Agentic AI Governance trail
- You assembled the stack. Who owns go-live?
- The unsexy AI: why your forklift matters more than your chatbot
- Adversarial evaluation for Agent Assist: ship agents that survive production
- How much guardrail does your AI agent need? What leaders must be able to defend
- Do you need enterprise AI orchestration? A 3-question readiness framework