A recent survey by OutSystems, titled 'The State of AI Development 2026', reveals that Artificial Intelligence (AI) has progressed into the early production phase for many enterprises, predominantly within the IT function. Based on responses from 1,879 IT leaders, the report issues a cautionary note: the rapid adoption of AI is at risk of outpacing essential governance and integration frameworks. A significant disconnect exists between what IT leaders envision agents accomplishing and their organizations' capacity for safe control. The authors strongly advocate for companies to establish robust controls and guardrails for AI systems, alongside emphasizing the crucial need to integrate new AI technologies seamlessly into existing organizational platforms.

The survey highlights widespread engagement with agentic strategies, with 97% of respondents exploring some form of this approach, and a notable 49% describing their current capabilities as 'advanced' or 'expert'. Nearly half of those surveyed reported that over 50% of their agentic AI projects have successfully transitioned from pilot to production. Indian companies emerged as leaders in successful implementation, with 50% indicating their AI projects are 51% to 75% successful. While 'cost reduction or efficiency gains' was the most frequently cited expectation for AI's impact, only 22% found their deployments most effective in this area. Instead, the most significant business gains were observed in empowering software developers with 'generative AI-assisted' tools.

The transition to AI agentic workflows is unevenly distributed across geographies and sectors. India stands out with the highest proportion of users considering themselves 'expert', whereas many organizations in Australia, Brazil, Germany, the Netherlands, the UK, and the US still categorize themselves as intermediate users. France and Germany exhibit the most skepticism regarding AI adoption, with Germany recording the highest share of leaders not utilizing agentic AI in any form. Financial services and technology sectors demonstrate the most significant movement from pilot to production, especially in core business functions, indicating a clear path from automation to measurable income returns. The report suggests that slower-moving sectors could benefit by emulating the implementation workflows of the fintech industry: commencing with narrow, high-volume tasks where performance is measurable and failures are containable, with a strong focus on the IT function.

Integration with legacy systems is identified as a critical factor for expanding agentic AI, with 48% of respondents citing it as the most important capability. Furthermore, 38% state that legacy systems are the primary cause for projects stalling between pilot and production. Overall, more than 40% of participants identified integration difficulties and legacy fragmentation as the most problematic barriers to AI development. The report challenges the notion that extensive data clean-up programs, often advocated by AI vendors as a prerequisite for successful deployments, are always necessary. It suggests that agents can be effectively built to operate in complex data environments, provided that governance and integration are simultaneously strengthened alongside AI implementation.

Despite these challenges, trust in agentic AI is on the rise. OutSystems reports that 73% of respondents express either high or moderate trust in allowing agents to operate autonomously, marking an approximate 10% increase from a similar survey conducted the previous year. Trust in code or workflows generated by third-party AI tools also saw a substantial increase, reaching 67% compared to only 40% who 'mostly trusted' generative AI for code writing without human intervention in the prior year. However, a significant governance gap persists: only 36% of respondents have a centralized approach to AI governance, while 64% lack such a facility, and 41% rely on rules implemented on a per-project basis.

The implementation of human-in-the-loop checkpoints presents technical difficulties for two-thirds of organizations, primarily due to the complex orchestration required to pause autonomous agents. Many organizations appear to be adopting looser oversight models, though it remains unclear if this stems from increased trust in AI models or pressure to deploy AI irrespective of security and reliability concerns. The report's authors caution that if this trend of reduced oversight continues, agentic AI adoption may advance more rapidly than the necessary methods of accountability. For firms aiming to scale agents in regulated or mission-critical environments, the survey findings emphasize the importance of treating orchestration and auditability – including logfiles and defined responsibilities – as integral components of the product itself.

A widespread concern highlighted is 'AI sprawl', although not explicitly defined, it is inferred to mean a lack of a centralized management platform for all enterprise AI deployments. A striking 94% of leaders expressed concern about this issue, with 39% being 'very' or 'extremely' concerned. Currently, only 12% of organizations utilize a centralized platform to manage and control this sprawl, underscoring a critical area for improvement in enterprise AI strategy.