Agents need governance
Enterprise AI is moving past the question of whether agents can do work to whether they can be controlled: firms struggle to prove AI ROI, and vendors are building governance layers like agent registries and managed agent workflows. A KPMG-backed analysis finds companies often can't measure baseline costs, while AWS has launched an Agent Registry and UiPath is packaging agentic ERP offerings to add permissions, reuse and auditability to deployments. (cio.com) (techzine.eu) (insidermonkey.com) (youtube.com)
Companies spent 2025 asking whether artificial intelligence agents could do real work, and they are spending 2026 asking a duller question that is turning out to be harder: who approved the agent, what systems can it touch, and how do you prove it saved money. (cio.com) A recent CIO report based on KPMG-backed analysis says many firms cannot calculate artificial intelligence return on investment because the work being replaced was never measured cleanly in the first place. If a company never knew the true cost of a finance handoff or support escalation, it cannot show exactly what an agent improved. (cio.com) That is why the market is shifting from “build another agent” to “build a system for managing agents.” Enterprises now want the same things they demanded for employees and software accounts: identity, permissions, reuse, logs, and a way to shut things off. (cio.com) (github.com) Amazon Web Services showed where this is going on April 9, 2026, when it put Agent Registry into preview inside Amazon Bedrock AgentCore. Amazon describes it as a private governed catalog for agents, tools, skills, Model Context Protocol servers, and custom resources inside an organization. (aws.amazon.com) A registry sounds boring, but it solves a very specific mess. When teams build agents in different departments, a registry becomes the company’s card catalog, so developers can find an existing agent, see who owns it, and reuse it instead of building a second version from scratch. (aws.amazon.com) (siliconangle.com) Amazon also says the registry is cloud-agnostic, which means it is meant to track agents even when they are not all running on Amazon’s own infrastructure. That matters because large companies rarely keep finance, customer service, and internal tools in one cloud or one vendor stack. (aws.amazon.com) (theregister.com) UiPath is pushing the same control-first idea from the workflow side instead of the cloud side. On March 12, 2026, UiPath said it was expanding its alliance with Deloitte to launch Agentic Enterprise Resource Planning, which packages agents with orchestration for complex back-office systems. (uipath.com) Enterprise resource planning software is the software that runs payroll, purchasing, inventory, and accounting, so mistakes there are expensive and visible. UiPath and Deloitte are selling agentic layers into that environment because companies want automation in the most rule-heavy parts of the business, but they also want approvals, traceability, and human review. (uipath.com) (deloitte.com) UiPath’s own platform already exposes the governance logic enterprises care about, including organization, tenant, service, folder, and project roles that determine who can access and manage workflows. In practice, that means agents are being treated less like chatbots and more like digital workers with job descriptions and badge access. (docs.uipath.com) The pattern across these launches is simple: companies are discovering that an autonomous tool is only useful if it can be audited like a financial system and constrained like an employee account. The next phase of enterprise artificial intelligence is not more impressive demos; it is registries, permissions, baselines, and logs that let a chief financial officer sign off on keeping the agents in production. (cio.com) (aws.amazon.com) (uipath.com)
