AI software development is rapidly moving from experimentation into production, but governance and integration challenges are becoming critical barriers to scale. A survey by OutSystems shows that 97% of organisations are exploring agentic AI strategies, with nearly half already moving projects to production environments.
Despite this momentum, the results highlight a clear gap between adoption and control. While cost reduction is often the primary goal, only 22% of organisations report achieving significant efficiency gains. Instead, the most measurable value comes from generative AI-assisted development, where developers use AI tools to improve productivity and accelerate coding workflows.
A major challenge is integration with existing systems. Nearly half of respondents identify legacy infrastructure as the largest barrier to scaling AI, with fragmented data and disconnected platforms slowing progress. However, the report suggests that organisations do not require perfect data environments to succeed. Instead, strengthening integration and governance alongside AI deployment can enable systems to operate effectively even in complex environments.
Governance remains the most significant risk. Only 36% of organisations have a centralised AI governance model; most rely on project-level controls. This fragmented approach increases the risk of “AI sprawl,” where multiple systems operate without oversight. At the same time, implementing human-in-the-loop controls is technically challenging, requiring orchestration systems that can pause or intervene in autonomous workflows.
- AI adoption is accelerating faster than governance capabilities
- Developer productivity is the strongest source of AI ROI
- Legacy integration remains a key barrier to scale
- Centralised governance is essential to control AI sprawl
Ultimately, AI software development success depends not just on deploying new tools, but on building centralised management and governance frameworks. By 2026, organisations that combine AI innovation with strong control and integration will be better positioned to scale safely and realise long-term value.
ソース：
