Why stalling may mean we are about to go faster
Companies learned that chatbots you can bolt onto existing systems are fundamentally different from AI agents that orchestrate workflows across CRM, supply chain, and finance operations.
I think ServiceNow's latest AI maturity data reveals something counterintuitive: the industry is finally getting serious about building agentic AI that actually works in enterprise environments, and the numbers prove it.
ServiceNow tracks how well organizations implement AI across their operations through an annual maturity assessment covering strategy, governance, and execution capabilities. This year's survey of over 4,500 global private and public sector leaders delivered unexpected results: average maturity scores dropped from 44 to 35 out of 100, with fewer than 1% of organizations scoring above 50. At first glance, this looks like AI adoption is stalling after the initial hype cycle. But the decline actually reflects organizations discovering the gap between generative AI pilots and production-grade agentic systems that execute real business processes. Companies learned that chatbots you can bolt onto existing systems are fundamentally different from AI agents that orchestrate workflows across CRM, supply chain, and finance operations.
The companies getting results are deploying 10+ specialized agents and treating exception handling as a core design requirement, not an afterthought. PwC found 83% of executives report faster problem-solving with agents, but only when they've built proper guardrails and human oversight into workflows that touch real business processes. The legal implications shift dramatically when AI moves from answering questions to making purchasing decisions or updating customer records.
For product counsel, this means rethinking data governance, liability allocation, and audit trails before the engineering team starts connecting agents to production systems. The organizations succeeding with multi-agent architectures are the ones who built sovereign data platforms first, which translates to having clear data ownership, processing agreements, and exception escalation procedures in place. When an AI agent flags a supply chain anomaly and triggers a reorder, someone needs to own the decision if it goes wrong.
The reset forces us to address the integration complexity we've been deferring. Smart money is on building the compliance infrastructure now while the technology is still settling into enterprise patterns. 🤖
Nicholas D. Evans
