In this guest blog post, Natalie Waller, Lead Product Marketing Manager at Fivetran, examines why agentic AI is stalling in most enterprises — not because of the AI itself, but because today’s data infrastructure wasn’t built for autonomous agents. Shifting to an Open Data Infrastructure (ODI) on Microsoft Azure gives organizations a practical path to scale AI workflows without accumulating the technical debt that’s sinking most pilots.
The promise of agentic AI feels like magic. Imagine a Microsoft Copilot shopping agent that doesn’t just suggest a product, but cross-references your past preferences, checks warehouse inventory, applies a personalized loyalty discount, and generates the checkout with pre-populated payment information from past purchases — all through a single natural language interaction. It’s an experience that feels human, but for that AI agent to act with true common sense, it requires a constant stream of data and context moving across your entire estate. For most companies, that "simple" interaction is currently impossible. Their infrastructure was designed for human analysts looking at yesterday's dashboards, not for autonomous agents that require continuous data freshness and clear definitions to avoid making critical errors.
Why "simple" AI is harder than it looks
Most AI pilots never make it to production, and the culprit is rarely the model — it’s the data. AI agents are hungry for data, but they need to be grounded in accurate, clearly defined business reality, not just what’s sitting in a shared folder. To complete a single transaction, an agent needs history, metadata, pricing, and inventory to work together instantly.
In the traditional enterprise, this can be a mess: Customer profiles sit in Microsoft Dynamics 365 or Salesforce, inventory levels are locked in legacy SAP or Oracle ERPs, and pricing logic is buried in a proprietary on-premises SQL database. To add more complication, critical context and problem-solving live within employee experience. When systems aren't synced and context isn’t available, the agent hallucinates. It confidently recommends a product that went out of stock three hours ago because its data foundation is brittle or disconnected. This isn't just a minor bug; it’s infrastructure debt that prevents AI from ever leaving the pilot phase.
From analytics to agent-ready
For years, data infrastructure was built around a single consumer: the human analyst. Dashboards, reports, and BI tools were the end goal, and batch pipelines that refreshed overnight were good enough.
Agentic AI changes the equation.
To move to production-scale AI, modern data stacks must evolve into Open Data Infrastructures: a model where data is portable, governed, and independent of any single vendor’s runtime. For data leaders, this isn’t a technology refresh; it’s a strategic shift in how the organization thinks about data ownership and accessibility.
The power of Open Data Infrastructure on Azure
This is where the partnership between Fivetran and Microsoft Azure removes the biggest barrier between AI ambition and AI reality. Available in Microsoft Marketplace, Fivetran eliminates the maintenance tax by replacing fragile, custom-coded pipelines with fully managed and automated movement from over 750 sources. Data lands in a Microsoft Fabric OneLake catalog in open Delta Lake formats, creating the unified foundation that Azure AI Foundry and Copilot Studio need.
As shown in the architecture above, Fivetran designed this flow to ensure your data is AI-ready from the moment it hits your lake:
- Extract and load: Fivetran moves data from 750+ sources, including high-throughput CDC from complex enterprise systems like SAP, Oracle, and mainframe environments directly into Microsoft Fabric.
- Transform: Through dbt integration and Fabric-trusted adapters, data is transformed into enriched, governed tables.
- Govern and store: Using Fivetran’s Managed Data Lake Service, data is stored in open Parquet/Delta Lake formats.
- What does this mean? The Managed Data Lake Service is a game-changer for data engineering teams because it makes a data lake "look and feel" like a high-performance data warehouse. It handles the complex plumbing, ACID compliance, schema evolution, and table maintenance so your engineers can stop managing files and start delivering data.
- Activate: This foundation powers real-time intelligence, Fabric data agents, and grounding for Microsoft Copilot.
This architecture solves the cost explosion associated with scaling AI. By utilizing Microsoft’s One Copy architecture, you land a single version of enterprise data that is immediately accessible to Azure Databricks, Azure Synapse Analytics, and SQL Database via shortcuts. This eliminates switching taxes, the hidden costs and compute spikes of moving data between different tools. It also removes the need for expensive data duplication. You store the data once and pay only for the compute you use, ensuring your infrastructure spend scales with business value, not just data volume.
What this looks like in practice
The value of this open architecture isn't theoretical. JetBlue moved from legacy, manual data processes to a centralized, automated foundation with Fivetran on Azure. By automating the movement of complex data into a unified environment, JetBlue can now set up new data pipelines in under two minutes — work that previously took engineers weeks, if not months, to build, test, and deploy. Today, its Snowflake environment contains more than 115TB of data from 130 systems, freeing data engineering teams to shift from "keeping the lights on" to high-value analytics and real-time operations that improve the passenger experience.
Interloop shows the same model in a Microsoft OneLake environment. With Fivetran-managed data movement, Interloop cut pipeline maintenance from 20 percent to 30 percent of analyst time to less than 5 percent and reduced analytics project turnaround time by more than 50 percent. Fivetran also helps lower Interloop’s Microsoft Fabric capacity costs by handling data extraction outside Fabric compute usage, giving Interloop and its clients more room to invest in analytics, AI, and customer-facing innovation.
Delivering the future of AI
The organizations that will win with AI are not necessarily those with the most sophisticated models; they’re the ones whose data infrastructure can support them. When your data foundation is automated and open, complexity stops being something you manage and starts being something you leverage.
By unifying your platform on Azure with Fivetran, you ensure your agents are grounded in the full context of your business reality, not a stale snapshot. Fivetran handles the heavy lifting of high-volume data movement, CDC, and schema evolution so you can focus on building AI experiences that work at scale. Ultimately, this allows your AI agents to operate on a complete, up-to-date view of the customer and every transaction.
Watch this webinar to see how Fivetran, Azure Data Lake Storage, Microsoft Fabric, and Copilot converge to create a scalable, governed data foundation. Want to learn more? Watch this demo or book a demo of Fivetran today.
