Despite Productivity Gains, Enterprise AI ROI Stalls: The 'Last-Mile' Gap and Agent Governance Challenge

The Fifth Annual Domino Enterprise AI Report, released today, reveals a paradoxical trend in enterprise AI adoption. While a substantial 93% of organizations report improved production capability in 2026, up from 88% in the previous year, a persistent 57% continue to find that their AI investments are not yielding a commensurate return. This marks the second consecutive year of this ROI plateau, signaling a significant hurdle in the journey towards AI maturity. The core of this challenge, as identified by the report, is the 'last-mile gap' – the chasm between AI models successfully deployed in production and the ability of business users to effectively act upon the insights generated. Thomas Robinson, COO at Domino Data Lab, emphasizes that merely deploying a model is no longer the ultimate milestone; true success hinges on the moment a business user can leverage AI findings at the pace and scale required by the business, within necessary governance frameworks. The report indicates a fragmented landscape for AI access, with 34% of organizations employing varied access methods across business units, and 40% still relying on mediated access, such as scheduled reports from data science teams. This situation underscores a broader trend where the initial excitement and investment in AI infrastructure and model development are now confronting the realities of operationalization and value realization. The rise of agentic AI, identified as a top organizational priority for enterprise AI leaders in 2026, further complicates this picture. While organizations are eager to deploy AI agents, the report notes a concerning lag in governance: almost as many organizations are running agentic AI without governance as with it, with 41% piloting or scaling agents without proper oversight. This mirrors earlier challenges faced in data governance and MLOps maturity, where technical capabilities outpaced the establishment of robust, secure, and compliant operational practices. The financial services sector, being heavily regulated, stands out in the report for leading in both governance maturity and production velocity, suggesting that a 'governance-first' approach can accelerate, rather than hinder, AI adoption. For cloud and DevOps practitioners, these findings mean a critical shift in focus. The role extends beyond deploying and managing AI infrastructure to actively bridging the 'last-mile gap.' This involves designing and implementing robust integration layers that seamlessly deliver AI-generated insights into existing business applications and workflows. Practitioners should prioritize developing user-friendly interfaces, APIs, and automation pipelines that enable business users to consume and act on AI outputs without requiring deep technical expertise. Furthermore, the rapid adoption of agentic AI necessitates a proactive stance on AI governance. This includes implementing comprehensive monitoring, auditing, and security frameworks specifically tailored for autonomous agents to ensure compliance, mitigate risks, and build trust. Organizations should invest in tools and processes that provide visibility into agent behavior, decision-making, and data interaction. The report implicitly calls for a more holistic MLOps strategy that encompasses not just model lifecycle management but also the entire value chain from data ingestion to actionable business outcomes, with governance embedded at every stage. Failure to address these operational and governance challenges will continue to impede the realization of AI's full transformative potential, making it imperative for technical teams to collaborate closely with business stakeholders to define clear success metrics beyond mere technical deployment.