Why AI Pilots Stall Before They Scale
Enterprises are pouring money into AI, but many are not seeing corresponding business outcomes. IBM’s CEO study found that only about a quarter of AI initiatives hit their expected ROI, and just 16% have reached enterprise-wide scale. Other analyst research echoes the same pattern: spending is rising quickly, while visible, measurable benefits remain concentrated in a small group of leaders. The root cause is less about models and more about operations. AI pilots are often built as isolated experiments, lacking a unified enterprise AI operating layer that connects agent orchestration, production data, monitoring, and governance. As projects move from demo to deployment, organizations hit bottlenecks around integration, risk controls, and ownership. Without a common fabric to manage agents, data flows, and policies, costs climb, complexity grows, and many promising proofs-of-concept stall before they can move into robust production.
IBM’s Enterprise AI Operating Layer: Agents, Data, Automation, Cloud
At its Think conference, IBM positioned an integrated operating layer as the missing link between AI pilots and production. The company frames successful AI at scale around four systems: agents, data, automation, and hybrid cloud. Its next-generation watsonx Orchestrate acts as an agent orchestration platform, described as an agentic control plane where organizations can deploy agents from multiple sources with consistent policies and accountability. IBM Bob, an agentic development partner, supports teams in designing these agents with built-in security and cost controls. On the data side, IBM is tying its Confluent acquisition to watsonx.data and Kafka- and Flink-based pipelines, plus a context layer that enforces governance at runtime and supports explainable decisions. IBM Concert extends the same operating model into infrastructure and security operations, correlating signals without forcing customers to rip and replace existing tools—key for enterprises wary of yet another siloed AI system.
Real-Time Data and Sovereignty as Core Design Requirements
For AI systems to move beyond pilots, they must operate on real-time data under strict, auditable controls. IBM’s integrated data layer aims to solve both dimensions. By connecting real-time event streaming with governed data stores and a semantic context layer, it enables AI agents to act on current operational signals while respecting enterprise AI governance rules. A proof of concept with Nestlé, spanning a global data mart across 186 countries, reported 83% cost savings and a 30x price-performance improvement, underscoring the operational upside of a unified layer. On the compliance front, IBM Sovereign Core targets regulated and cross-border workloads, embedding policy at the infrastructure runtime level. Capabilities such as customer-operated control, in-boundary identity, encryption, continuous compliance monitoring, audit evidence, and governed AI execution are designed to make sovereignty and compliance intrinsic to the platform—no longer add-ons bolted on after pilots succeed.
Xurrent: Making AI the Fabric of IT Service and Operations
While IBM targets broad enterprise AI, Xurrent focuses its operating layer on IT service and operations management. The platform has long embedded AI through Sera AI, which classifies requests, drafts knowledge articles, and resolves routine tickets, with most customers already using it in production. Its new autonomous AI Agents go further: they act as digital team members, completing tickets end-to-end across the IT service lifecycle, with humans setting guardrails and approving work when necessary. Every action is logged and governed by the same policies that apply to human staff, aligning tightly with enterprise AI governance requirements. Xurrent’s Shared Policy and Data Layer gives all agents—vendor-built or customer-built—the same governed view of the IT environment. Rather than treating AI as a bolt-on feature, Xurrent positions AI as the fabric of its cloud-based platform, blending productivity gains with a complete audit trail and consistent controls.
Open Standards, MCP, and the Future of Agent Orchestration
As enterprises adopt more AI agents, interoperability becomes as critical as intelligence. Open standards like the Model Context Protocol (MCP) are emerging to ensure that agents can interact with diverse models and systems while remaining subject to a unified governance layer. Xurrent’s open MCP server illustrates this shift: it allows customers to plug in external AI models from any provider or from in-house development. Once connected, those models inherit Xurrent’s Shared Policy and Data Layer, gaining the same governance, audit trail, and visibility as native agents. This creates an enterprise AI operating layer where service requests, operational data, and multiple AI providers converge under one control plane. Combined with IBM’s focus on agent orchestration, real-time data, and sovereignty, the message is clear: the next phase of AI is not about isolated pilots, but about standardized, governed platforms where agents can safely operate at scale.