What 1,900 IT leaders know about scaling agentic AI

OutSystems surveyed 1,900 IT leaders worldwide for our first state of AI development report. This is a departure from our past research. For almost a decade, we’ve released a state of application development report, and it’s always been well received. However, today, AI is assisting developers with everything from generating requirements and writing code to testing and optimization. It’s becoming so integral to software development that focusing on just applications is no longer sufficient. So, the report’s horizons have been broadened.

In our 2026 survey, IT leaders answered questions about their AI strategies, use, plans, and concerns. One of the top topics was governance. In most AI conversations, it’s framed as a bottleneck; however, the data says the opposite is true. Organizations with centralized governance over their AI development environment are iterating 58% faster than those that have not.

Read on to learn more about current challenges with agentic AI, what IT leaders are doing to overcome them, how governance is critical to their success, and six approaches that separate AI programs that scale from those that stall.

Or, if you want to dive into all the goodness, you can go straight to the state of AI development 2026 report.

Most enterprises are building on a foundation that doesn’t scale

The report shows that 97% of organizations are actively exploring agentic AI, and 96% are already running agents in some capacity. One retail C-level executive described what early progress looks like on the ground. "We've found agentic AI most useful when it quietly removes friction by tidying workflows, removing manual effort, and giving teams more time for the work that actually matters. There's still caution around security, but the practical benefits are becoming hard to ignore."

That kind of impact is promising. Yet, only 49% of respondents have moved more than half of their agentic AI projects from pilot to full enterprise production. Generally, the tendency is to assume that reasons are talent, budget, or executive commitment. Our data says it’s existing systems and architecture.

Most organizations are not choosing between old systems and new agents. They’re trying to run both simultaneously, and this is a struggle. Forty-eight percent say integration with existing enterprise infrastructure is the single most important capability they still need to expand their programs. Yet, 38% say legacy systems are the primary reason their AI projects have stalled because the ability to scale just isn’t there.

The AI sprawl gap is wider than most organizations realize

AI sprawl is a top concern, with 94% of respondents saying they’re concerned about it, and 39% describe themselves as very or extremely concerned about it. And yet only 12% use a centralized platform to actually maintain control over it.

That 82-point gap between concern and action reflects a structural challenge. Most organizations are governing their agentic AI programs based on project-level rules or ad hoc approaches, and 41% rely on these piecemeal tactics as their primary governance mechanism. Unfortunately, these approaches leave gaps in security, compliance, and visibility that grow wider with every new agent deployed.

The organizations inside that 12% are seeing something different. Among teams that have centralized their AI governance platform, 58% report faster iteration and delivery, 48% report easier governance and policy enforcement, and 45% report shared data context across their systems. Those are not marginal improvements. They are the difference between an AI program that accelerates the business and one that multiplies the complexity of managing it.

A senior manager in the UK government described the balance that makes this work. "Agentic AI can work independently to solve problems faster and improve efficiency. It can help organizations make better decisions by analyzing data and taking action with little human help. However, it is important that agentic AI is used responsibly and is well-supervised."

That combination of speed and oversight is exactly what centralized governance makes structurally possible, rather than something each team has to engineer separately.

IT is where enterprise AI earns the right to scale

The realized ROI from agentic AI is currently concentrated in technology departments, with 40% of value captured so far coming from IT development productivity. Operational efficiency accounts for 22%, and customer experience adds another 14%. Revenue generation and sales sit at 11%. Although front-office use cases that appear in most AI roadmaps are still aspirational for the majority of organizations, the IT pattern is one that scales.

IT operations and data analysis are the two most-explored use cases in the survey, cited by 55% and 52% of respondents respectively. Teams using agents to automate manual IT workflows, reduce development cycle friction, and extract consistent value from data are doing something more strategic than improving their own productivity. They’re building the governance models, trust levels, and architecture patterns that will carry the same approach into sales, HR, customer support, and beyond.

A senior manager in financial services saw exactly this trajectory developing. "The integration of agentic AI in workflows and data management has improved our productivity with minimal input. The development and use of better agentic AI in the future will have great impact in finance and data management."

The governance model a CIO establishes in IT operations today becomes the template every other function inherits tomorrow.

Governance is an architectural decision

Building human-in-the-loop checkpoints for AI is technically difficult, according to 66% of the IT leaders surveyed. It requires an orchestration layer that can pause agents mid-task, preserve their reasoning context, and produce decision logs for human review. Without that infrastructure, organizations default to passive monitoring.

This challenge compounds when development and governance tools are not integrated from the start. When agents are built in one system, deployed through another, and monitored by a third, the orchestration layer becomes another integration project. The technical debt accumulates before the agent reaches production.

The leaders addressing this most effectively are treating it as an architectural choice. Nearly all leaders surveyed view a unified platform for building and managing apps, data, and AI agents as very or somewhat important for their future. The capabilities they prioritize reflect what the technical challenge actually demands. Legacy system integration comes first at 48%, followed by centralized monitoring and control at 47%, multi-model flexibility without vendor lock-in at 41%, and built-in governance and auditing at 37%.

A senior director in Australia framed the organizational reality behind those requirements. "Agentic AI's success depends on strong governance, integration with existing systems, and ongoing support for adoption across teams."

Six approaches that separate AI programs that scale from those that stall

The report findings point toward a clear sequence for moving from AI experimentation into enterprise production. These are approaches that IT leaders across regions and industries consistently identified as foundational.

1. Start with full portfolio visibility

Establish a system for monitoring every agent and application in your environment that is designed to evolve as adoption matures.

2. Identify immediate use cases and prove value quickly

Prototype in high-friction areas of IT operations, demonstrate the impact, and use those results to build the organizational confidence required for broader rollout. Axos Bank is using the OutSystems platform to build an agent that reverse-engineers the legacy modernization process, allowing them to gain control over old custom apps. It analyzes code to provide details about these apps. The team also built an intelligent log analysis agent to interpret error logs and provide recommendations, eliminating the need for manual analysis.

3. Harmonize your existing data before you scale your agents

Agents perform significantly better when they have access to consistent, well-governed data across your legacy systems and custom applications. Data fragmentation compounds every other challenge on this list.

4. Track costs at the token level from the beginning

Monitor what each agent consumes, understand the cost per token against your AI provider contracts, and measure that against the outcomes each agent delivers. This discipline becomes critical at enterprise scale.

5. Build human-in-the-loop flows deliberately

Identify the workflows in your portfolio where human oversight is required and engineer those checkpoints from the start, with explicit accountability for every decision agents initiate. For example, leading European media company TX Group deployed an AI agent that analyzes invoice data, suggesting human intervention when appropriate.

6. Embed security and access controls in every agent before deployment

Role-based permissions, usage limits, and lifecycle governance should not be layered on after the fact. They should be part of how every agent is built.

The before and after that defines this moment in enterprise AI

Most IT leaders reading this report recognize this "before" state:

• Multiple development tools operating independently
• Inconsistent monitoring across agent deployments
• Project-level governance rules applied unevenly
• Agents that function in silos from the enterprise systems and the data that would make them most valuable

Each new capability added to the portfolio increases the surface area of risk without adding the visibility to manage it.

The after state is what 96% of respondents say they want, and what 7% have already adopted. A unified environment where agents are built alongside the enterprise applications they interact with, governed through consistent controls, and managed across their full lifecycle without requiring a separate stack for each phase of development.

OutSystems was designed specifically to bridge that gap. Our AI development platform brings agent building, AI-assisted app generation, enterprise data integration, and full lifecycle governance into a single environment that organizations already trust for mission-critical application development.

"OutSystems is the enterprise platform that provides the governance and trust layer we need to have on top of agentic AI-led development. It allows us to benefit from the speed of AI in a controlled and secure way, ultimately making it possible to use AI without spending hours reviewing the code behind its output," says Cristiano Marques, Director of Software Development, at Vopak, the world’s leading independent tank storage company.

Dive into the details

Read the full state of AI development report for complete industry and regional breakdowns. You can also watch our webinar or request a personalized demo to see how OutSystems manages the full AI development lifecycle. Or, start building AI agents today with a free OutSystems personal environment.