Artificial Intelligence in Production: Between Enthusiasm and Control Challenges

A survey conducted by OutSystems, titled "The State of AI Development 2026," highlights how artificial intelligence has now reached the early production phase in numerous enterprises, with a predominant implementation within IT functions. Based on the responses of 1,879 IT leaders, the research issues a warning: the speed of AI adoption risks outpacing corporate governance and integration capabilities. This discrepancy creates a gap between IT leaders' expectations regarding AI agent functionalities and their organizations' ability to manage them securely.

The report's authors urge companies to strengthen controls and "guardrails" on AI systems, while also emphasizing the crucial importance of integrating new AI technologies into existing platforms. Nearly all respondents (97%) are exploring agentic strategies, with 49% describing their current capabilities as "advanced" or "expert." Approximately half of the surveyed companies have already moved over 50% of agentic AI projects from pilot to production, with India standing out for the highest success rate in implementation.

The Key Role of Developers and Integration Challenges

While cost reduction and efficiency gains are the most cited expectations for AI, only 22% of companies found the greatest benefits in these areas. Instead, the most effective area for business gains proved to be equipping software developers with generative AI-assisted tools. This suggests that the most lasting value from agentic AI initially manifests internally, supporting the productivity of development teams rather than directly in customer interactions.

Integration with legacy systems emerges as one of the main challenges. 48% of respondents consider integration with existing infrastructures as the most important capability needed to expand agentic AI, and 38% indicate legacy systems as the primary reason projects stall between pilot and production. Over 40% cited integration difficulties and data fragmentation as major obstacles. Contrary to what is often argued by some vendors, the report suggests that massive data clean-up programs may not always be necessary, as agents can operate effectively in complex data environments, provided that governance and integration are strengthened in parallel with AI implementation.

Governance, Trust, and the Risk of "AI Sprawl"

Trust in agentic AI is, however, improving. OutSystems reports that 73% of respondents express either high or moderate trust in allowing agents to act autonomously, an increase of approximately 10% compared to the previous year. Trust in code or workflows generated by third-party AI tools is slightly lower, at 67%, but still represents a substantial increase from 40% the previous year.

Despite this growing trust, only 36% of companies adopt a centralized approach to AI governance, while 64% lack such a facility, and 41% rely on rules implemented on a per-project basis. Two-thirds of respondents find it technically difficult to implement "human-in-the-loop checkpoints," as they require complex orchestration capable of pausing agents, effectively inserting a manual brake on potentially autonomous operations. This trend towards looser oversight models could accelerate agentic AI adoption but raises questions about accountability mechanisms.

Implications for On-Premise Deployment and Data Sovereignty

The lack of centralized governance is a widespread concern: 94% of leaders are alarmed by the phenomenon of "AI sprawl," understood as an uncontrolled proliferation of AI implementations without a unified management platform. 39% are very or extremely concerned, but only 12% currently use a centralized platform to contain this phenomenon.

For companies looking to scale the use of agents in regulated or mission-critical settings, the survey findings underscore the importance of considering orchestration and auditability as an integral part of the product. In scenarios where compliance is paramount, traceability through detailed logfiles and clear definition of responsibilities are essential elements for any agentic AI rollout. This is particularly relevant for organizations evaluating on-premise Deployment, where direct control over infrastructure and data can facilitate adherence to data sovereignty and compliance requirements. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate the trade-offs between control, security, and Total Cost of Ownership (TCO) in these scenarios, providing tools for informed decisions without specific recommendations.