Chloe Maraina is a visionary in the realm of business intelligence, driven by a deep-seated passion for transforming massive, complex datasets into compelling visual narratives. With a background that seamlessly blends data science with high-level data management, she has become a leading voice in how organizations integrate and leverage proprietary information to fuel the next generation of artificial intelligence. Today, she joins us to discuss the shifting tides of AI development, exploring how the industry is moving away from the “plumbing” of data integration and toward a more agile, governed framework for building context-aware agents.
Our discussion centers on the critical evolution of developer tools that bridge the gap between stagnant enterprise data and dynamic AI models. We examine the transition from traditional integration methods to a governed runtime data layer, the impact of standardized protocols on developer productivity, and the strategic importance of creating a “developer-first” ecosystem. Chloe also provides insights into why specialized connectivity layers are becoming the essential architectural control points for enterprises that are serious about moving their AI projects out of the lab and into production.
Development teams frequently report that a staggering amount of their time is consumed by the “plumbing” of data integration—nearly 75 percent, according to some reports—rather than actually building or training models. From your perspective, why has this remained such a persistent hurdle for AI innovation?
The reality is that enterprise data is incredibly messy and siloed, which creates a massive friction point for anyone trying to build something as sophisticated as a context-aware AI tool. When we look at the results of the study commissioned in late 2025, which surveyed more than 200 data and AI leaders, it’s clear that the impetus for new tools was born out of frustration with this exact bottleneck. Developers are often forced to go through long, arduous enterprise procurement cycles just to get access to the proprietary data they need, which kills the momentum of any project. Instead of spending their days refining model logic or improving accuracy, these highly skilled professionals are stuck writing custom connectors and navigating complex API schemas. By the time they actually get the data flowing, the business requirements might have already shifted, making the 75 percent of time spent on integration feel like a lost investment.
There has been a lot of buzz surrounding the transition of companies from being simple “connectors” to becoming “governed runtime data layers.” How does this architectural shift actually change the way a developer interacts with enterprise systems?
This shift is truly transformative because it moves the data layer directly into the environment where the developer is already working. Instead of having to build a bridge to the data every single time, the data becomes a standardized, accessible layer that AI agents can consume directly and in real-time. By exposing enterprise APIs as a unified data layer with standardized schemas and full read/write support, we are effectively removing the need for IT to be involved at every single step of the process. This creates a much more fluid workflow where developers can call upon the data they need through a single URL or a command-line interface. It turns what used to be a month-long integration project into a simple API call, allowing for the kind of consistent development that is necessary to scale AI across a large organization.
The introduction of the Connect AI Developer Edition and the Python SDK seems to be a direct response to a “developer-first” mentality. How do these tools specifically lower the barrier to entry for teams working with frameworks like LangChain or assistants like Claude Code?
The most significant advantage here is that these tools are designed to meet developers exactly where they live, whether that’s in a terminal or a Python script. The Developer Edition is particularly impactful because it provides the full enterprise feature set for free, allowing teams to prototype and iterate without worrying about immediate costs or administrative red tape. For someone using LangChain or Cursor, the ability to import governed data directly into their Python workflow without changing a single line of their existing code is a game-changer. It’s all about reducing the “cognitive load” of data management; when you can package governed access into a single Model Context Protocol server URL, the AI agent only calls on the specific data it needs. This precision not only makes the development process faster but also ensures that the resulting AI tools are much more contextually aware and reliable.
With major players like Snowflake and Databricks building massive, all-encompassing platforms, how do you see more specialized, focused connectivity layers maintaining their relevance in the enterprise stack?
While the giants of the industry are focused on building “everything-and-the-kitchen-sink” platforms to handle the entire AI lifecycle, there is a distinct and growing need for specialized layers that focus purely on the architectural control points of connectivity. Companies like CData aren’t necessarily trying to outbuild the massive storage and compute capabilities of a Snowflake; instead, they are carving out a niche by owning the trusted connection between AI intelligence and operational systems. Every enterprise architecture needs a way to bridge the gap between the brain—the LLM—and the hands—the enterprise data sources. By focusing narrowly on this connection and offering one of the largest libraries of enterprise connectors available, specialized providers become the essential glue that keeps those larger platforms functional. It’s a strategic play that prioritizes flexibility and deep integration over the “walled garden” approach.
As AI agents become more autonomous and start handling more sensitive production data, the conversation inevitably turns to trust and oversight. What are the specific governance features that will determine whether an agent is truly “production-ready”?
We are moving into an era where “good enough” is no longer acceptable for enterprise AI, especially as agents move toward higher levels of autonomy. For an agent to be trusted near production data, there must be a robust framework of audit trails, strict access policies, and end-to-end encryption that aligns with organizational guidelines. We are seeing a major investment in improving the quality of semantic models to ensure that when an agent accesses data, it truly understands the business logic and context behind it. Without these “guardrails,” an autonomous agent could easily make decisions based on misinterpreted data, which is a risk most enterprise teams are not willing to take. The future of the ecosystem depends on advancing platform maturity so that every action an agent takes is transparent, compliant, and fully reversible if necessary.
What is your forecast for the evolution of the “semantic layer” in enterprise AI over the next few years?
I anticipate that the semantic layer will evolve from a static map of data into a dynamic, “living” domain-specific model that proactively guides AI agents through complex enterprise tasks. We will likely see the rise of pretrained semantic layers specifically tailored for high-stakes functions like customer experience or financial reporting, which will drastically cut down the time it takes to deploy specialized agents. Beyond just providing context, I expect these layers to incorporate advanced observability features that show exactly how AI models are utilizing data in real-time. This level of visibility will be the “holy grail” for organizations, allowing them to not only optimize the performance of their models but also to identify entirely new opportunities for automation that were previously hidden in the noise of their data. As these semantic layers become more intelligent, the friction between raw data and actionable AI will eventually vanish, making the 75 percent integration hurdle a relic of the past.
