OutSystems' latest survey of 1,879 IT leaders reveals a striking disconnect in enterprise AI adoption: while 97% of organizations are pursuing agentic AI strategies, the technology is delivering value in unexpected places. Companies anticipated cost reduction and efficiency gains, but the real returns are materializing inside IT departments, specifically at developers' desks.
The data shows 49% of surveyed organizations rate their agentic AI capabilities as "advanced" or "expert," with nearly half reporting that over 50% of their AI projects have transitioned from pilot to production. Yet only 22% found their deployments most effective at reducing costs. Instead, 40% report the strongest ROI from equipping software developers with generative AI-assisted tools—nearly double the returns seen in operational efficiency.
Why Developer Tools Are Winning the AI Race
This outcome makes sense when you examine the risk profile and feedback loops involved. Software development environments offer controlled settings where AI-generated code can be tested, reviewed, and rolled back without customer impact. Developers can immediately assess whether an AI suggestion works, creating a tight iteration cycle that accelerates learning for both the human and the machine.
Compare this to customer-facing deployments, where errors carry reputational risk and require more sophisticated orchestration. The survey indicates IT operations (55%) and data analysis (52%) lead as the most explored use cases, while customer experience trails at 33%. Organizations are following a sensible path: prove the technology works internally before exposing it to external stakeholders.
Financial services and technology sectors are furthest ahead in this transition, moving AI implementations into core business functions. These industries benefit from high-volume, repeatable workflows where performance metrics are clear and failures can be isolated. A trading algorithm or fraud detection system generates measurable outcomes that justify continued investment. Other sectors would do well to study this playbook rather than attempting to leapfrog directly to customer-facing applications.
The Governance Gap That Could Derail Progress
The survey exposes a troubling vulnerability: 64% of organizations lack centralized AI governance, with 41% relying on ad-hoc, per-project rules. Only 36% have implemented a unified governance framework. This fragmentation becomes critical as trust in autonomous AI agents rises—73% of respondents now express high or moderate trust in letting agents act independently, up roughly 10 percentage points from the previous year.
Rising trust might sound positive, but it creates a dangerous scenario when governance lags behind. Two-thirds of respondents say building human-in-the-loop checkpoints is technically difficult because it requires orchestration systems capable of pausing autonomous operations. Without these controls, organizations face a choice between slowing AI adoption or accepting risks they may not fully understand.
The concern about "AI sprawl" reflects this anxiety—94% of leaders worry about it, with 39% very or extremely concerned. Yet only 12% currently use a centralized platform to manage AI deployments across the enterprise. This suggests many organizations are deploying AI faster than they can track it, creating shadow AI systems that operate outside formal oversight.
Legacy Systems: Barrier or Scapegoat?
Conventional wisdom holds that enterprises must clean up legacy systems and consolidate data before deploying AI at scale. The OutSystems survey challenges this assumption. While 48% cite integration with legacy systems as crucial for expanding agentic AI, and 38% blame legacy systems for stalled projects, the report's authors argue that agents can function effectively in complex data environments if governance and integration are strengthened alongside implementation.
This perspective shifts the conversation from expensive, multi-year data modernization programs to more targeted integration work. Rather than waiting for a pristine data environment that may never arrive, organizations can deploy AI agents that navigate messy systems—provided they invest in the orchestration layer that connects these agents to existing platforms.
The practical implication: IT leaders should allocate resources to integration frameworks and governance structures rather than postponing AI projects until legacy systems are replaced. The technology can handle fragmentation; the question is whether organizations can handle the governance complexity that comes with it.
Geographic and Sectoral Divides
India stands out as the most advanced market, with 50% of Indian companies reporting that 51-75% of their AI projects succeed in production. Indian organizations also claim the highest share of "expert" users. Meanwhile, Germany records the highest percentage of leaders not using agentic AI at all, with France and Germany showing the most skepticism overall.
These geographic patterns likely reflect regulatory environments, risk tolerance, and the maturity of local tech ecosystems. Organizations in Australia, Brazil, Germany, the Netherlands, the UK, and the US predominantly identify as intermediate-stage users, suggesting a middle ground between aggressive adoption and outright resistance.
The sectoral data reinforces that financial services and technology companies are setting the pace. These industries have clear metrics for measuring AI impact and can afford the experimentation required to refine implementations. Healthcare, manufacturing, and retail organizations may need different approaches that account for regulatory constraints and operational complexity.
What IT Leaders Should Do Now
The survey data points to three immediate priorities. First, focus AI investments on internal IT functions where returns are proven rather than chasing customer-facing applications that require more mature governance. Developer productivity tools offer the fastest path to measurable ROI and build organizational confidence in the technology.
Second, establish centralized governance before AI sprawl becomes unmanageable. This doesn't mean halting deployments—it means creating visibility into where AI operates, what decisions it makes, and how humans can intervene when necessary. The 12% of organizations using centralized platforms are positioning themselves to scale safely; the other 88% are accumulating technical debt.
Third, treat orchestration and auditability as core product requirements, not afterthoughts. In regulated industries or mission-critical settings, the ability to pause an agent, review its decision-making process, and maintain audit trails isn't optional. Organizations that build these capabilities now will move faster later, while those that skip this step will face costly retrofits or compliance failures.
The transition from AI pilots to production is happening faster than many expected, but success requires matching technical capability with organizational readiness. The companies that balance innovation with governance will capture the value; those that prioritize speed over control may find themselves managing crises instead of celebrating wins.
