The Context Layer: Enterprise IT's Next Foundational Investment
Why the maturity of your data foundations and not your AI tooling will decide whether agentic AI pays back
At Axtria Ignite 2026, two senior pharma data leaders from Takeda and Novartis, told a room filled with enterprise data leaders that the decade-long investment in clean master data was necessary but no longer sufficient. The next layer of enterprise architecture is the context layer: the structure that lets language models reason over both structured and unstructured data the way the business actually operates.
“A true data product is going to need to require all this kind of meta-information wrapped around it that's very potentially language-based and not numerical in nature.”
A Layer the Industry Keeps Mislabeling
Semantic layer, context layer, knowledge graph, tribal brain - treating these as synonyms is actively slowing down enterprise programs.
“The semantic layer is basically you creating an asset on top of your actual data, your metadata, your business context, business glossaries, your business rules. When you say context layer, now you have to combine all of them and create a context out of it.”
- Semantic layer: Curatorial work - harvesting and defining metadata, glossaries, and business rules into a reusable asset.
- Context layer: Integrative work - joining those assets to structured and unstructured data so an LLM can reason over both.
The Cost Risk Your Board Will Ask About
There's a real cost risk enterprise IT leaders should carry into every agentic AI business case.
“If you design it inefficiently, you could find a situation where the cost of running your agents is potentially higher than the cost of an in-house capability center or a global capability center. So it's not a free trip.”
The anchor case was a mature data and AI platform supporting neuroscience brands, where an AI layer performed well specifically because years of curated data and controlled prompting sat underneath it before the AI was ever added.
“It took so long to build the trust and maturity of that underlying data set to then put this piece on top.”
Key Takeaways
- The context layer is a distinct architectural investment from the semantic layer. Budgeting for one does not fund the other.
- Custody, not ownership, is the correct operating model for central IT. Brands own the logic, IT translates and governs it.
- An inefficient agent architecture can cost more than the operation it was meant to replace. You need to model this explicitly before funding a build.
- The foundational maturity of your data estate and not the sophistication of the AI tooling determines whether the context layer pays back this fiscal year or next.
FAQs
