Chloe Maraina is a visionary in the realm of business intelligence, known for her unique ability to transform massive datasets into vivid, actionable narratives. As an expert in data science and enterprise integration, she has spent years helping organizations bridge the gap between abstract algorithmic potential and concrete business results. Her perspective is particularly vital today, as the industry shifts from simple chatbots to sophisticated agentic systems that can navigate complex corporate ecosystems.
In this discussion, we explore the evolving landscape of enterprise AI, focusing on how organizations can finally realize a return on their technological investments. We delve into the mechanics of natural language “Skills” that allow AI to move beyond mere conversation into actual task execution, and we examine the critical role of unified governance in managing multi-agent networks. We also touch upon the practicalities of modern software development, the importance of transparency in AI reasoning, and the future of data as the central control plane for the entire enterprise.
Many organizations struggle to see a return on AI investments because models often lack specific business context. How do natural language “Skills” and workflow execution address this gap, and what metrics should teams track to verify that agents are actually completing work rather than just answering questions?
The frustration is palpable across the industry right now, especially when you realize that a staggering 95% of organizations haven’t seen a real return on their AI investments yet. The bottleneck isn’t the intelligence of the models themselves, but rather the “context gap” where the AI simply doesn’t understand the specific levers of a particular business. By introducing “Skills,” we are essentially giving the AI a handbook on how to actually do the job—it moves from being a passive observer to an active participant that can execute workflows in natural language. Instead of just tracking “query volume” or “response accuracy,” teams need to look at “task completion rates” within third-party systems like Salesforce or Jira. When an agent can autonomously navigate a complex business process and produce a tangible “Artifact”—like a finished report or a synchronized calendar event—that is when you know the investment is finally paying off.
Integrating data from third-party systems like AWS Glue or Databricks while using communication tools like Slack and Salesforce often creates fragmented workflows. What are the practical steps for building a multi-agent network across these external systems, and how does a unified governance layer change the way developers manage these pipelines?
Building a multi-agent network starts with moving away from the “walled garden” mentality and embracing protocols like the Model Context Protocol (MCP) and Agent Communication Protocol (ACP). These connectors act as the connective tissue, allowing an agent to pull data from an AWS Glue environment and push an update into a Slack channel without a human having to manually stitch the tools together. The real magic, however, happens at the governance layer, which acts as a centralized command center where security and permissions are applied uniformly across the entire pipeline. For a developer, this is a massive relief because it means they don’t have to rebuild security protocols for every individual integration. You gain this sense of “controlled freedom” where agents can interact with Databricks or Google Docs, but always within the guardrails of enterprise-grade security that keeps the proprietary data safe.
Trusting an autonomous agent requires understanding its multi-step reasoning across structured and unstructured data. How do features like automated logic reports and shared artifacts help teams validate an agent’s output, and what specific anecdotes have you seen where this transparency helped prevent a costly error in production?
Trust is earned through transparency, and in the world of data, that means being able to see exactly how an agent “thought” through a problem. Automated logic reports are essentially a transcript of the agent’s reasoning, showing how it pulled from unstructured documents and structured tables to reach a conclusion. I’ve seen instances where an agent was tasked with supply chain optimization and, because of these logic reports, the human supervisor noticed the agent was over-relying on an outdated PDF instead of a live data feed. This transparency allowed the team to course-correct the workflow before a massive, incorrect order was placed, saving the company from a significant financial headache. When you can share these “Artifacts” and visualizations across a team, the AI stops being a “black box” and starts feeling like a reliable, highly diligent colleague whose work you can actually peer-review.
Moving between local development environments like VS Code and cloud-native platforms can slow down the deployment of AI applications. How do new software development kits and platform-specific plugins accelerate the transition from experimentation to production, and what is the typical timeline for a team to operationalize a new agentic system?
There is often a jarring “context switch” when a developer has to leave their favorite IDE, like VS Code, to mess around in a cloud console, and that friction kills momentum. By providing native plugins and a dedicated Agent SDK, we allow developers to build, test, and orchestrate their agents within the environment where they are most productive. This integration can shave weeks off the development cycle because you’re no longer “stitching together” disparate tools; you’re building within a cohesive ecosystem. While every project is different, we are seeing teams move from a raw concept to a functional, operationalized agentic system in a fraction of the time it used to take—often moving through the experimentation phase in just a few weeks. The goal is to make the transition into production feel like a natural extension of the development process rather than a grueling architectural overhaul.
AI agents are increasingly expected to learn from user interactions to deliver personalized responses over time. What are the technical trade-offs between maintaining static governance and a system that adapts its workflows based on user behavior, and how should enterprises balance this personalization with strict data security requirements?
The tension between personalization and security is one of the most delicate balancing acts in modern data management. On one hand, you want a system that learns from how a specific manager uses data—adapting its visualizations and insights to their unique preferences over time—but you can’t let that learning process bypass your core security rules. The technical solution lies in “continuous learning” modules that sit on top of a governed foundation, ensuring the agent adapts its delivery but never its access rights. You create a personalized experience by allowing the agent to remember user preferences and past interactions, which makes the tool feel intuitive and “smart,” while the static governance layer remains the ultimate arbiter of what data can actually be touched. It’s about creating a system that feels like a personal assistant but behaves like a disciplined security officer, ensuring that even the most personalized response never leaks sensitive information.
What is your forecast for the future of agentic AI as a control plane for enterprise data?
I believe we are moving toward a future where data itself becomes the primary control plane for the entire enterprise, acting as the central nervous system that directs every AI agent and automated workflow. Instead of having dozens of fragmented applications, companies will operate through a unified, governed data foundation where agents have the business context needed to make high-stakes decisions autonomously. We will see a shift away from “vendor lock-in” as interoperability becomes the gold standard, allowing organizations to swap models and tools seamlessly while their data remains the stable, protected core of the operation. Ultimately, the successful enterprise of the future won’t just “use” AI; it will be orchestrated by a network of intelligent agents that understand the business as well as any human executive, all powered by a single, transparent source of truth.
