Chloe Maraina is a powerhouse in the world of Business Intelligence, known for her ability to transform massive datasets into clear, actionable visual narratives. With a deep foundation in data science and a forward-thinking approach to integration, she has spent her career helping organizations navigate the complex intersection of raw data and strategic decision-making. Her expertise is particularly relevant as enterprises shift from simple AI experimentation to the rigorous demands of production-ready applications. Today, she joins us to discuss the recent technological shifts at Qlik, focusing on how the gap between AI ambition and actual readiness is being bridged through agentic data engineering, high-trust data products, and a renewed focus on governance.
In this interview, we explore the systemic challenges that cause AI pilots to fail, specifically the data engineering bottleneck that prevents models from reaching production. We discuss the introduction of automated data quality agents that utilize trust scores and anomaly detection to ensure information integrity. The conversation also covers the evolution of data reusability through the creation of governed data products, the importance of interoperable architectures like the Model Context Protocol, and the upcoming strategic shift under new leadership as the industry looks toward the next phase of operational AI.
The move from AI pilots to full-scale production is notoriously difficult, with many projects stalling before they ever deliver value. From your perspective, what is the primary cause of this friction, and how are modern data engineering tools finally addressing the gap between ambition and reality?
The bottleneck is truly felt at the engineering layer, where the sheer weight of manual data preparation can crush even the most ambitious AI initiatives. Organizations often find themselves in a state of paralysis because the data informing their agents is incomplete, outdated, or simply incorrect. When an AI agent is forced to make inferences based on low-quality information, the resulting outputs can lead to devastating consequences, including significant lost revenue and dangerous regulatory noncompliance. By embedding agentic AI throughout the data engineering lifecycle, we are finally seeing tools that allow for the discovery, validation, and packaging of trusted data products at a pace that matches business needs. This shift moves us away from simply using AI to generate snippets of code and toward a holistic system where the data itself is governed and “AI-ready” the moment it is called upon.
Data quality is often cited as the “silent killer” of AI projects. How do the new automated agents and metrics like trust scores change the way a data engineer interacts with their pipelines on a daily basis?
In the past, data engineers had to manually hunt for anomalies, often finding errors only after they had already poisoned a report or an AI model. The introduction of specific agents for data quality allows users to create and edit rigorous quality rules that run autonomously in the background. These agents measure the health of data points and entire datasets, providing tangible trust scores that give stakeholders immediate confidence in the information they are consuming. When the system detects an anomaly, it doesn’t just sit there; it reports the issue instantly, allowing for a proactive response rather than a reactive cleanup. This level of automation means engineers are no longer bogged down by the minutiae of data cleaning and can focus on the strategic architecture of the multi-agent networks they are trying to build.
There has been a lot of discussion regarding the shift from traditional data marts to the concept of reusable “Data Products.” Why is this distinction so important for companies trying to scale their AI and analytics efforts without duplicating work?
For years, the industry struggled with data reusability, often falling back on creating isolated data marts stored in proprietary files that were difficult to adapt for new purposes. The transition to governed Data Products allows teams to build and manage assets that are intentionally designed to be operationalized and reused across multiple AI and business use cases. Instead of recreating an entire dataset every time a new initiative pops up, organizations can now establish a reliable foundation that serves as a single source of truth for the entire enterprise. This approach not only reduces the massive engineering backlogs that plague IT departments but also ensures that every project, whether it’s a simple dashboard or a complex predictive model, is built on the same high-quality, governed foundation.
As organizations integrate more third-party AI assistants and proprietary business logic, security and interoperability become major concerns. How does the Model Context Protocol help protect sensitive information while still giving AI the context it needs to be effective?
One of the biggest hurdles for any enterprise is allowing an AI to be “smart” without exposing the “crown jewels” of proprietary data to the open web. Expanded Model Context Protocol capabilities solve this by enabling authorized agents to access business logic and sensitive data within a strictly secure environment. This gives the AI the specific context it needs to perform—such as understanding internal terminology or unique customer segments—without compromising the overall governance or lineage of the data. It is a delicate balance of providing enough information for the AI to be useful while maintaining the controls that enterprises require for safety. This open architecture approach is vital because it allows customers to work with their preferred technology stack rather than being locked into a single, restrictive ecosystem.
The leadership landscape at major tech firms often signals a shift in strategy, and with a new CEO stepping in this summer, how do you see the vision of data management evolving under fresh guidance?
Leadership changes, such as the sudden resignation of Mike Capone who led since January 2018, often mark the beginning of a more mature phase in a company’s evolution. Saugata Saha, who officially begins his role on July 31, brings a wealth of experience from the market intelligence sector that will likely sharpen the focus on how data is consumed and monetized. We are moving away from the era of “AI for the sake of AI” and into a period where operational reliability and business context are the primary drivers of development. The goal now is to strengthen the data foundation so that the transition from experimentation to production is seamless. This means we will likely see more tools that help users combine disparate capabilities across a single, governed platform, making the entire workflow more efficient and less fragmented.
Looking toward the future, specifically into late 2026, there is a growing interest in AI observability. Why is it becoming necessary to monitor not just the data, but the actual interactions between users and AI agents?
As we deploy more complex multi-agent networks, simply knowing that the data is “clean” isn’t enough; we need to understand the entire lifecycle of an AI’s decision-making process. Future systems will likely need to monitor agent-question interactions, consumption patterns, and usage frequency to ensure that the outputs remain accurate over time. By tracking these patterns, we can identify exactly where a pipeline might be failing or where an AI agent might be making incorrect inferences based on shifting data trends. This level of observability allows for a complete feedback loop, where we can assess the impact of data quality issues on final business outcomes. It turns the “black box” of AI into a transparent, auditable process that can be refined and improved with every single interaction.
What is your forecast for the role of the data engineer over the next few years as AI agents become more autonomous?
I believe we are entering an era where the data engineer will transition from being a manual “builder” of pipelines to an “orchestrator” of intelligent systems. By 2026, the manual grunt work of detecting anomalies and standardizing terminology will be almost entirely handled by agentic workflows, leaving the humans to focus on the high-level architecture and the ethical governance of the data. We will see a significant reduction in the time it takes to move from a pilot to a production-ready application, as the bottleneck of data readiness is finally cleared through automation. Ultimately, the successful organizations will be those that treat their data as a living, reusable product rather than a static asset, allowing them to pivot quickly as new AI capabilities emerge.
