Written by: Moatassim (Mo) Aidrus

We're in the fourth year of the GenAI adoption wave, and 2026 is predicted as the critical year for enterprise AI ROI. There's a clear pattern: companies deploy ChatGPT, Claude AI, or build custom LLM applications. They get solid results on generic tasks. Then they hit a wall trying to use these tools for anything that requires understanding their actual business.

The problem isn't the models. It's context. Or more precisely, the complete absence of structured, queryable, machine-readable context about how an enterprise actually works.

Most discussions about enterprise AI focus on RAG patterns, vector databases, and prompt engineering. These are important, but they're solving the wrong problem first. Before you can retrieve context, you need to actually have context: documented, structured, and maintained. And most organizations simply don't.

Think about what an LLM needs to know to actually help your business:

  • What your data entities are and how they relate
  • What your business terms mean and how they're calculated
  • Where data comes from and how it transforms
  • What's reliable and what's not
  • What the business logic actually is behind your metrics

This information exists, but it's locked in structures that weren't designed to serve AI applications. Your data warehouse schema, transformation code, BI definitions, data catalog. The context is there, just not in formats AI systems can consume.

The Unstructured-Structured Gap

Here's where the rubber meets the road. Enterprise AI tools have proven they can handle unstructured data. Companies like Glean have built significant businesses making documents, wikis, Slack conversations, and emails queryable. The value is real and customers are paying for it.

But the moment business users get comfortable with AI-powered search, they ask the inevitable question: "Can we also get insights from our data warehouse?" Revenue trends. Customer churn rates. Product profitability. The numbers that actually run the business.

And this is where sales pipelines get clogged. The answer is technically yes, but practically no. Because those insights require semantic models. Entity relationships, business logic, semantic meaning, data lineage, aggregation rules. The structured knowledge layer that tells AI what "revenue" means in your business, how customers connect to orders, what aggregations are valid.

Creating these semantic models manually is why traditional analytics consulting engagements run 3-6 months and cost $1M+ or more. Requirements gathering, data modeling, YAML specifications, DBT code development, testing, deployment. It's necessary work, but it's also a complete blocker.

So what happens? The enterprise AI vendor can't expand into structured data without solving a problem outside their core competency. The sale stalls or gets downsized. The customer's data warehouse AI capabilities sit unused. Everyone loses.

Meanwhile, Snowflake and Databricks have built their own AI layers. Cortex Agents. Genie. Powerful capabilities that should complete the picture. Except they hit the exact same prerequisite: semantic models. Without that context layer, the AI tools are like a sports car without fuel.

The Missing Layer

Data warehouses and data catalogs are among the major sources where enterprise context naturally accumulates. They already integrate with your systems, they have the metadata, they understand relationships and lineage. But that context is trapped in formats meant for humans using UIs, not for AI systems making decisions in real-time.

The gap in the market isn't another RAG framework or vector database. It's the layer that automatically generates, maintains, and serves enterprise context in formats that AI applications can actually consume. Something that can look at your data warehouse, understand the structure, extract the business logic, map the relationships, generate the documentation, and expose all of that as queryable knowledge.

We're solving exactly this with Ekai. We connect to your data warehouse, automatically generate ERDs and understand relationships, capture business context through our semantic modeling, and produce all the artifacts that downstream AI applications need: data catalogs, business glossaries, metrics definitions, lineage maps, validation rules. The entire context layer, generated and maintained automatically. And we can serve it.

For Snowflake, this means going from business requirements to deployed Cortex Agents in hours instead of months. We're available as a Snowflake Native App, so all processing happens within your Snowflake account. Your data never leaves your environment.

For enterprise AI tools handling unstructured data, this means their customers can finally get the structured data insights they've been asking for. The semantic modeling blocker gets automated away.

2026: The Year Context Gets Solved

The companies that will win with enterprise AI this year aren't necessarily the ones with the best models or the most sophisticated prompts. They'll be the ones who solved the context problem. Companies like Ekai that will build the infrastructure to capture, maintain, and serve their enterprise knowledge in ways that AI systems can actually use.

The technology exists. The platforms are ready. The missing piece is the automation to connect them.

author credits: along with my co-founder Hussnain Ahmed

Related: The Nike Reset: How Elliott Hill Is Turning Strategy, Marketing, and Culture Into Growth