Much of the agentic AI conversation over the past year has focused on what agents can do. A quieter but more telling shift is happening around who manages them once they're deployed at scale — and two major platform launches this year show how fast that problem has grown.

Microsoft's answer: a control plane for agents

Microsoft's Agent 365 went generally available on May 1, 2026, positioned explicitly as a "control plane" for governing, securing, and managing the growing number of AI agents operating inside an organisation. It's priced at $15 per user per month, and also bundled into Microsoft 365 E7 — pricing that signals Microsoft expects agent governance to become a standard line item in enterprise IT budgets, not a niche add-on.

OpenAI's answer — and its own platform churn

OpenAI took a parallel but different approach, launching Frontier in February 2026, an enterprise platform for building, deploying, and managing agents across business software, with early adopters including Uber, Intuit, State Farm, and Thermo Fisher Scientific. What's notable is what OpenAI is retiring alongside it: the company's earlier AgentKit, Agent Builder, and Evals tools will become unavailable from November 30, 2026, with developers redirected to Frontier and the Agents SDK instead. For a set of tools launched only months earlier, that's a fast replacement cycle — and a reminder that the tooling layer around agentic AI is still far from settled.

Why this creates a new kind of job

Enterprise surveys on agent adoption vary widely — vendor-sponsored research from CrewAI and OutSystems puts current usage at 65-97% of surveyed organisations, while independent telemetry from security firm Snyk found only about one in five organisations actually running autonomous agents in production. That gap matters: it suggests a lot of agentic AI activity today is piloting and experimentation rather than settled, production-grade deployment — which is precisely the stage where governance, access control, and oversight roles get created, before processes mature and headcount around them stabilises.

In India, that pattern is already showing up in hiring language. Industry coverage of enterprise AI rollouts has pointed to new titles like "agentic engineer" emerging specifically to manage and govern deployed agents, distinct from the engineers who build them. An EY survey of enterprise leaders found that while nearly half report running multiple live agentic AI use cases, only about 10% have reached enterprise-wide deployment — the gap between those two numbers is roughly where governance roles tend to get created.

What it means if you're job hunting

• "Agent governance," "AI agent operations," and "agent management" are worth watching as emerging job-posting language, distinct from AI/ML engineering roles.
• Familiarity with access control, audit logging, and compliance frameworks is becoming relevant to AI roles in a way it wasn't two years ago.
• Expect the tooling to keep shifting — platform churn like OpenAI's AgentKit retirement means staying tied to one vendor's agent-building tools is riskier than building transferable skills in agent oversight and evaluation.