主なポイント
- Most enterprises still run data and integration teams as two separate tracks, an approach that breaks down the moment agentic AI needs real-time, unified data.
- The new O’Reilly report Beyond Pipelines: The New Data Operating Model for AI at Scale, sponsored by Boomi, unpacks the hidden costs of data fragmentation and the principles behind an AI-ready data foundation.
- Organizations that unify their data early build a compounding advantage — AI velocity, the combination of speed, scale, and trust that lets autonomous agents operate safely and effectively.
Most enterprises still run two parallel tracks for their data: data teams building pipelines into a warehouse or lakehouse, and integration teams connecting applications for business processes. It’s an old divide, built around tools that used to require very different skills. These siloed team operations led to the introduction of the DevOps model back in 2009 and DataOps in 2014. Today’s platforms have closed that technical gap, but most organizations’ team structures haven’t caught up.
The gap didn’t matter much when the data was being used to create a static report or even a dashboard. It matters enormously now that AI agents are expected to take action autonomously or semi-autonomously.
A new O’Reilly report, “Beyond Pipelines: The New Data Operating Model for AI at Scale,” written by data strategist Adam Morton and sponsored by Boomi, digs into exactly why that gap is now a business risk, and what leaders need to build instead. If you’re currently evaluating your data and integration strategy for AI, or exploring how to govern AI agents responsibly at scale, this asset is a useful place to start. This post shares a glimpse into what you’ll learn in our preview release of the first two chapters.
Fragmented Data Prevents AI Readiness
Data resides in multiple locations within every organization. Customer data might live in Salesforce, while financials and HR are in different SAP modules. Purchase orders might be in NetSuite, product data is in a database somewhere, and marketing data is in Marketo. For organizations with storefronts, things get more complex with the addition of systems like Shopify and inventory management tools. For years, that fragmentation was inefficient but workable, because data teams and integration teams could use the data they worked with, as needed, to fulfill their own requirements — reporting on one hand, transactional workflows on the other.
Agentic AI breaks that arrangement. Agents run on data. Better data produces better agents, which produce better customer experiences, which generate more high-quality data. The report calls this loop the AI velocity flywheel. And if you don’t provide your agents with reliable, validated data — accessible in real-time (or at least close to it) — you’re setting them up to fail.
Three Principles for an AI-Ready Data Foundation
Rather than prescribing a specific tech stack, the report argues for an architectural philosophy built on three principles:
- Strategic centralization over universal movement: Centralize only the data assets that need it, and reach the rest through unified interfaces, instead of copying everything into one warehouse.
- Live state over historical snapshots: Streaming and event-driven integration in place of scheduled ETL, so agents act on current reality instead of yesterday’s export.
- Embedded intelligence over separate systems: AI built into the data infrastructure itself (connectors, cataloging, quality checks) rather than bolted on downstream.
One important caveat: centralizing data doesn’t mean centralizing control. Governance still needs to be addressed. A well-designed, unified access layer can enforce permissions automatically to ensure compliance, allowing all your AI agents to operate without every action requiring human review and approval. The next two chapters of Beyond Pipelines will dive deeper into roles, governance, and team structure, and what changes might be needed to better support AI success.
Ready to move beyond pipelines? Download “From System Failure to System Future” to get started.