AI investment is no longer the constraint. Execution is.
With global AI spending projected to reach $2.5 trillion, the conversation has shifted from experimentation to operationalization. The real challenge facing enterprises today is not whether AI can deliver value, but whether organizations can govern, scale, and trust it across complex, distributed environments.
In this episode of AppDevANGLE, I spoke with Harsha Kumar, CEO of NewRocket, about what it actually takes to move from pilot projects to production-grade agentic AI—particularly within enterprise platforms like ServiceNow.
The discussion makes one thing clear: AI success is no longer about model capability. It is about data readiness, governance frameworks, and the ability to operationalize AI consistently across the business.
Agentic AI Moves Beyond Automation Into Decision-Making
Traditional enterprise automation has always been deterministic. It follows predefined rules, executes repetitive tasks, and relies on humans for judgment when something breaks.
Agentic AI fundamentally changes that model. “What this means is… agentic AI can actually apply reasoning and take a lot of conditional steps in workflows similar to what a human would do,” said Kumar.
Instead of simply executing tasks, agentic systems interpret context, evaluate options, and take action across multi-step workflows. They can escalate decisions when needed, but increasingly handle the bulk of operational work on their own.
This shift moves automation from task execution to outcome ownership. It also raises the stakes. The more autonomy these systems have, the more important it becomes to understand how decisions are made and controlled.
Governance Becomes the Control Plane for AI
As agents begin operating across systems of record, governance is no longer a compliance checkbox; it becomes the core control layer of the architecture. “You can’t have agents just running amok. You need identity, access control, privacy, and tracking of the transactions being done,” Kumar explained.
This introduces a new requirement for enterprises: governance must be embedded directly into AI workflows, not layered on afterward. Organizations need to know what agents are doing, what data they are accessing, and what outcomes they are producing.
NewRocket’s introduction of its Maestro platform, working alongside ServiceNow’s AI Control Tower, reflects this broader shift. The goal is not just visibility, but accountability: tracking decisions, enforcing policies, and tying activity back to measurable business value. Without this layer, AI doesn’t scale; it fragments.
Data Readiness Remains the Primary Barrier
For all the progress in AI tooling, the biggest blocker to production remains unchanged: data. “Clean data, enough of it, is the fuel for AI,” Kumar said.
The issue is not simply having data. It is having data that is consistent, contextual, and usable across systems. Large enterprises often struggle with fragmented datasets, inconsistent definitions, and unclear semantics.
A simple example illustrates the problem: even something as basic as a “due date” can mean different things across systems—end of day, start of day, or completion deadline. These inconsistencies compound quickly at scale.
What emerges is a reality where AI performance is tightly coupled to data quality. If the data is incomplete or inconsistent, the output will be unreliable. As a result, data preparation is not a one-time step, but rather, an ongoing operational discipline.
Systems of Engagement Introduce New Risk
As enterprises move toward agentic AI, they are also shifting from isolated systems of record to interconnected systems of engagement. That transition introduces a new layer of risk.
Historically, data governance benefited from containment. Systems were siloed, and access was tightly controlled. AI changes that dynamic by extracting and acting on data across environments.
This raises new questions around data lineage, access control, and compliance. Once data leaves its source system, maintaining governance becomes significantly more complex.
To address this, modern architectures are evolving toward models where data can be accessed without being duplicated, and where every interaction is tracked. Whether through zero-copy data access or unified data fabrics, the goal is to maintain control without limiting flexibility.
“All of this has to be tracked… otherwise, it can be absolute chaos if a transaction goes wrong,” Kumar noted.
Execution Discipline Separates Leaders From Laggards
The gap between organizations that successfully operationalize AI and those that remain stuck in pilots is growing. The difference is not ambition. It is execution.
Organizations that are making progress tend to start with a clear understanding of the business outcome they are trying to achieve. They align their platforms to support that outcome, invest in data quality, and build governance into the process from the beginning.
“Having a clear outcome in mind… and a roadmap to build towards that is super critical,” Kumar said.
In contrast, organizations that approach AI as a series of disconnected experiments often fail to move beyond proof-of-concept. Without a defined value model and operational structure, even promising pilots stall.
AI Adoption Is an Organizational Shift
One of the more understated themes in the discussion is that AI transformation is not just technical but also organizational.
Enterprises need to rethink how work is structured, how decisions are made, and who is responsible for building and maintaining these systems. The rise of agentic workflows is blurring the line between developers, operators, and business users. “Bring in change agents… people who are more AI-native,” Kumar advised.
This reflects a broader trend: AI is lowering technical barriers while increasing the importance of domain expertise and adaptability. Success depends less on specialized skills and more on the ability to integrate AI into real-world processes.
Analyst Take
Agentic AI is reshaping enterprise application development by shifting software from passive tools to active participants in decision-making.
This transition introduces a new operational model. Data becomes the foundation, governance becomes the control plane, and agents become the execution layer. Each of these elements must work together for AI to deliver value at scale.
The most important takeaway is this: AI is no longer a feature. It is an operational system. Organizations that treat it as an isolated capability will continue to struggle with fragmentation and stalled pilots. Those that approach it as a coordinated transformation that is grounded in clean data, embedded governance, and clear business outcomes will be the ones that successfully operationalize AI.
The next phase of enterprise AI will not be defined by better models. It will be defined by better systems.
