Survey: 76% of tech managers face scrutiny on AI-enhanced Agile ROI

Artificial intelligence is reshaping how software development teams work, particularly within Agile methodologies that have dominated the industry for over two decades. While Agile promised to align technology teams more closely with business needs, many organizations have struggled to achieve this vision at scale. Now, AI tools and autonomous agents are emerging as potential catalysts to finally unlock Agile’s full potential—though not without significant risks.

This shift comes at a critical moment. Technology leaders face mounting pressure to demonstrate clear return on investment from their software projects, while traditional Agile practices have remained largely stagnant for nearly a decade. According to Digital.ai, a software delivery platform company, in their recent survey of nearly 350 Agile coaches and consultants from large enterprises, the majority of organizations still describe their Agile implementation as “better than nothing but could be more effective.”

The survey reveals a troubling disconnect: 76% of technology managers report increased scrutiny on the business impact of their Agile initiatives, yet only 49% have established proper guardrails as AI adoption accelerates. This gap between AI enthusiasm and governance creates both unprecedented opportunities and substantial risks for software development teams.

The current state of Agile development

Agile methodology, which emphasizes iterative development and close collaboration between business stakeholders and technology teams, has hit a plateau. For nearly a decade, surveys have consistently shown that only 52% to 60% of organizations describe themselves as “using Agile practices but still maturing,” with just one in ten reporting high competency levels across their entire organization.

This stagnation persists despite growing business demands for faster, more effective software delivery. The methodology that once promised to revolutionize software development now finds itself “damned with faint praise at the exact moment when organizations need it most,” as the Digital.ai report notes.

Enter AI-powered development tools. Unlike previous technological advances that simply automated existing processes, AI agents can potentially reason, decide, and act autonomously throughout the software delivery lifecycle. This represents a fundamental shift from AI as a supportive tool to AI as an active participant in development workflows.

However, industry experts urge caution about overly optimistic expectations. “AI is not something you can pull out of your toolbox and expect magical things to happen,” warns Andrew Kum-Seun, research director at Info-Tech Research Group, a technology research and advisory firm. “IT managers must be prepared to address the human, workflow, and technical implications that naturally come with AI while being honest about what AI can do today for their organization.”

6 essential guidelines for AI-enhanced Agile development

1. Acknowledge the experimental nature

Current AI implementation in software development remains largely experimental, particularly for autonomous activities. Kum-Seun reports that his firm has yet to observe widespread adoption of AI agents performing fully independent tasks in the software development lifecycle (SDLC)—the complete process from initial planning through deployment and maintenance.

Most successful implementations today focus on individual, task-level support where AI complements human team members rather than replacing them. This reality check helps set appropriate expectations and prevents organizations from overcommitting to unproven approaches.

2. Begin with low-risk pilots and comprehensive preparation

Smart AI integration starts small and methodical. “Pilot low-risk processes like test generation and documentation before exposing live code or customer data,” advises Zbyněk Sopuch, chief technology officer of Safetica, a data security company.

Test generation involves AI creating automated tests to verify that software functions correctly, while documentation refers to AI writing explanations of how code works—both relatively safe applications that don’t directly impact customer-facing systems.

Before implementing any AI tools, organizations must map their entire data landscape, including source code repositories, system logs, support tickets, and customer information. This inventory helps determine which data should remain off-limits to AI systems and establishes clear boundaries for safe experimentation.

The recommended approach follows a “govern first, then scale” philosophy: establish comprehensive policies defining acceptable AI use before expanding tools or user access across the organization.

3. Identify specific improvement opportunities

AI shows particular promise in several key areas of software development management. Backlog analysis—the process of reviewing and prioritizing pending work items—becomes more sophisticated when AI can sort tasks by business impact and surface critical priorities, allowing developers to focus on the most pressing issues.

Smart test creation represents another significant opportunity. AI can write test cases directly tied to specific requirements or previous software bugs, improving both coverage and traceability. This means teams can better ensure their software works as intended while maintaining clear connections between tests and business needs.

Documentation traditionally consumes significant developer time but benefits enormously from AI assistance. AI can automatically draft pull request summaries (explanations of proposed code changes), user notes, and release updates based on code version history, freeing developers for more complex work.

AI also excels at identifying bottlenecks in software delivery pipelines, helping teams recognize and address workflow problems more quickly. Additionally, AI-driven dashboards can translate technical metrics into business-friendly summaries, enabling non-technical stakeholders to better understand and engage with development processes.

4. Maintain human oversight and explainability

Since Agile methodology emphasizes human collaboration and communication, AI implementation must support rather than undermine these principles. Leaders who fail to bridge their AI vision with team concerns risk seeing improper tool usage or even deliberate implementation sabotage.

Successful AI integration requires explainable systems that clearly cite the sources of their suggestions. “No black-box code that cannot be simply verified,” emphasizes Sopuch. Teams need AI tools that show their reasoning, not mysterious recommendations that developers cannot understand or validate.

Human approval remains mandatory for critical decisions. While AI can write and refactor code, humans must approve all merges (integration of new code into the main codebase), product releases, and any exceptions to standard processes. Every AI interaction should be logged, including prompts, outputs, and human approvals, enabling comprehensive audits when needed.

5. Respect fundamental software development principles

Despite technological advances, core software development challenges persist. “I have seen many silver bullets, like Agile and AI, which have come and gone over the years,” observes Laura Zuber, training and customer support manager for Quantitative Software Management, a software analytics firm. “Some have measurably improved software development productivity. However, the fundamentals of software development remain the same. It is challenging.”

Organizations must expect initial productivity decreases when implementing AI tools, similar to any significant process change. Productivity gains emerge only after developers become proficient at prompting AI systems and teaching them to understand specific project needs. Developers need sufficient experience to recognize poor or irrelevant AI-generated code and continue working with AI agents until they produce acceptable results.

This learning curve requires patience and realistic timeline expectations from business stakeholders who may expect immediate improvements.

6. Understand and mitigate data exposure risks

The primary risk in AI-enhanced Agile development mirrors other AI initiatives: potential data exposure. Developers can accidentally include sensitive information in prompts sent to external AI tools, such as API keys (authentication credentials), passwords, or customer data. These inadvertent disclosures can create serious security vulnerabilities.

Shadow AI—unauthorized use of AI tools by employees seeking productivity improvements—poses additional risks. When team members install or use unapproved AI applications without IT visibility, organizations lose control over data governance and compliance requirements. This grassroots adoption, while often well-intentioned, can create significant regulatory and security exposures.

Organizations need clear policies defining acceptable AI tools, data handling procedures, and approval processes for new applications. Regular training helps developers understand these boundaries and recognize potential risks in their daily workflows.

The broader implications

This AI-driven transformation differs fundamentally from previous software development evolutions like waterfall methodology, DevOps practices, cloud automation, or mobile development. According to the Digital.ai analysis, “Agentic AI is not a new tool; it’s a new teammate. It brings awareness and adaptability to every stage of delivery, creating systems that learn from each interaction and continuously improve flow, quality, and security.”

This characterization suggests a future where AI agents actively participate in software development decisions rather than simply executing predefined tasks. However, realizing this vision requires careful balance between technological capability and human judgment, particularly in maintaining the collaborative spirit that makes Agile methodology effective.

For business leaders evaluating AI integration in their software development processes, success depends on realistic expectations, comprehensive preparation, and commitment to gradual, measured implementation. The technology holds genuine promise for addressing long-standing Agile challenges, but only when deployed thoughtfully within well-governed frameworks that preserve human oversight and decision-making authority.

The organizations that navigate this transition successfully will likely gain significant competitive advantages in software delivery speed and quality. However, those that rush into AI adoption without proper preparation risk creating new problems that could set back their development capabilities rather than advancing them.