For years, enterprises have built intelligence in a set order. The data gets integrated and moved to a central platform. After that, it has to be normalized and governed. Someone builds reporting models, and only then do people start getting value from it. There are still good reasons to work this way. Large-scale analytics needs it, and so do historical analysis and regulatory reporting, along with any organization that wants one governed analytical data model.
There’s another route that has opened up now, thanks to Model Context Protocol (MCP).
AI doesn’t have to wait until every bit of information has been moved somewhere new. It can be given governed access to the systems where the information already lives. On the surface, that’s a minor shift. The consequences are bigger than they sound.
From ‘moving data’ to ‘accessing capabilities’
Most established organizations already have the information they need. It’s just scattered across dozens of applications, sometimes hundreds. Finance sees one piece. HR sees another. Operational systems have their own share. So do the CRM, the service desk tools, document repositories and any legacy applications built for that industry. Getting the full picture usually meant building integration pipelines and copying it all into a warehouse, lake, or lakehouse.
MCP changes the question from:
“How do we get all of this data into one place?”
to:
“How do we allow an AI system to securely reach the information and capabilities it needs, wherever they already exist?”
Under an MCP-based architecture, an organization chooses which data and functions in its existing applications AI can access. The applications themselves don’t get rebuilt or replaced. A legacy system is then more than a data source sitting in the migration queue. It’s a live enterprise capability an AI agent can interact with.
A virtual intelligence layer across the enterprise
Imagine asking:
“How are staffing levels, service demand and spending changing across our operating units this month?”
Right now someone would likely export data from a few systems, line it up in spreadsheets and work out what the numbers mean. Once governed MCP connections exist, an AI agent could, in principle, go straight to the authorized systems and ask. It would match the answers on the spot and return a report, a chart, or a plain explanation. The data itself may not have moved. The intelligence layer moved to the data.
The source architecture describes this kind of experience. Authorized data comes in from multiple operational systems and lands in one report or graph. Each figure still links back to its source record. For many operational questions, this can work like a virtual data layer, without first needing a wholesale enterprise data migration.
Reporting is only half the story
This gets more interesting when the architecture can act, not just read. Business intelligence tells you what’s happening. An agentic system might act on it.
A manager might request:
“Offboard this employee.”
The agent wouldn’t just write up a to-do list. With the right authorization, it could look up the person’s accounts and assets and make the allowed changes in HR, identity, email and equipment systems. Anything odd gets flagged, and anything sensitive would stop and request human approval first.
So the reporting connections you set up can also take action across the enterprise. That’s a much bigger deal than swapping dashboards for chat. People get, and later AI agents too, one shared layer for asking questions, working through them, and acting on the systems they already run.
Why this matters for legacy modernization
Modernization has usually forced a choice. One option was to leave the aging applications alone and live with information spread across them. The other was to pay for big programs to replace applications, migrate data and build new analytical infrastructure. MCP might open a third path: modernize the interaction layer first, then bring the systems underneath later.
A system that’s awkward to hook up to modern AI doesn’t need replacing just for that. Instead, you can expose selected capabilities through governed interfaces, and later use cases can reuse those connections. Maybe the first one answers a reporting question. The next might run an automated workflow, and another might back an AI assistant. Each new use adds value to what’s already there. That is why MCP could matter a lot in government, healthcare, financial services, and anywhere else core systems stay in production for decades.
Does this eliminate the lakehouse?
No. A modern data platform still earns its keep, and that matters. Big analytical workloads still need one. So do long-term historical analysis and keeping definitions consistent across the whole enterprise. Advanced machine learning, high-performance aggregation, and regulatory-grade consolidated datasets fall into the same category.
Organizations may find that some AI use cases can move forward without that infrastructure. Operational intelligence, cross-system questions, and agentic workflows can often be built directly on top of governed enterprise systems.
That can radically change the modernization sequence.
Instead of:
Centralize → Transform → Analyze → Automate
organizations can increasingly pursue:
Connect → Govern → Ask → Act
and centralize data later where the economics and use case justify it.
Leapfrogging infrastructure
The most interesting thing about MCP may have little to do with the protocol. It opens up architectural leapfrogging. Some organizations spent the past decade building huge centralized data environments before they deployed any intelligent applications. Organizations with large legacy estates may not have to take that same route. Instead, they may be able to set a governed AI layer on top of what they already own and start getting joined-up answers straight away.
Put another way, nobody should have to move everything before they can use anything. First connect the systems that exist. Then limit what AI can see and do. Where the consequences matter, add human approval. Start with a single question worth answering, and take action once you have more confidence. That’s the same progression the source architecture follows.
Enterprise data architecture doesn’t go away. Organizations can separate the timeline for becoming AI-enabled from the timeline for replacing or consolidating their technology estate. For organizations with decades of legacy investment, that may be one of MCP’s most important consequences.
To understand more about how our perspectives translate into practical solutions :
Data + AI | Enterprise Applications | Cloud Readiness for AI.
Author Profile:
Michael Kares is a technology and enterprise sales leader with three decades of experience helping organizations navigate complex technology landscapes and drive digital transformation through practical, scalable solutions.