MIT reports that 85% of organizations want to be agentic within the next three years. The catch: 76% admit their current operations and infrastructure aren't ready for the shift. 

Snowflake and Databricks are betting their next decade on closing that gap. 

The message at both of their summits was nearly identical: Agentic AI systems are ready to run inside your enterprise, and they're the vendors who can make it happen. Snowflake framed their pitch around the AI Data Cloud, while Databricks pushed Unity AI Gateway and Genie. 

Whether these promises hold up in production is yet to be seen. Still, these are the changes Snowflake and Databricks customers should look out for in the second half of 2026.

What Snowflake is betting on: More layers, same walls

In 2026, Snowflake plans to extend the AI Data Cloud from a place where data lives to a place where agents work. Here are the summit announcements that support that shift: 

Introducing Snowflake Cortex Sense

Cortex sits at the core of Snowflake's agentic intelligence play, and the suite is adding layers faster than most teams can track them. 

Until this summit, there were four: Cortex Code, Cortex Analyst, Cortex Search, and Cortex Agents. This year, Code became CoCo (more on that below), and a fifth layer joined the lineup: Co rtex Sense. Essentially, Cortex Sense is a context layer that feeds Snowflake's AI agents with business definitions and operational knowledge.

The direction is right. AI context makes agents reliable, and a dedicated layer for it is exactly the kind of investment most teams are interested in. But Sense doesn't stand alone. It sits on top of Horizon Catalog, which means the business meaning it feeds to agents is only as good as the definitions you've already engineered into Horizon. Every new metric, every renamed field, every business rule that changes has to flow back into Horizon before Sense can pass it on to an agent reliably.

Also, Cortex Sense is still in private preview, untested against the genuinely messy, undocumented, historically accumulated data estates that characterize most enterprise environments.

Advancing Horizon Catalog

Snowflake repositioned Horizon Catalog from a metadata store to a semantic layer. The thinking behind the move is straightforward: Every agent action and conversational query should start from a shared, governed definition of business metrics. It's Snowflake's answer to the perpetual debate over inconsistent answers and accuracy. However, this answer only works if your definition lives in Snowflake. 

The outcome for your team splits two ways. Horizon-native teams get an upgrade that works in their favor. Everyone else is being asked to migrate their metrics and their operational context into Horizon, or accept that agents will draw from a partial picture.

Acquiring Natoma 

This is arguably one of Snowflake’s most strategic moves in 2026. Natoma built the enterprise MCP (Model Context Protocol) gateway—a centralized registry that lets AI agents connect securely to business applications, APIs, and databases. With this acquisition, Snowflake is positioning itself as the control plane for how agents connect to enterprise systems. 

Here’s what you should know: The gateway routes agent requests. Whether or not the returning answers are reliable is still an open question. That’s a problem the gateway was not designed to address.

Rebranding Snowflake Intelligence with CoCo and CoWork 

Snowflake just dropped a fresh rebrand on two of its flagship AI products: 

Together, they add an AI layer on top of your warehouse data 

The new names take some getting used to, but there’s a bigger challenge at play. These tools are both unable to natively pull from cross-system data and tools, meaning your governance, metrics, definitions, and data that sit outside of Snowflake still aren’t accessible.

Partnering with Anthropic

Another big announcement from Snowflake Summit is the partnership with Anthropic to power CoWork and CoCo. It’s borrowing Anthropic's enterprise-trust reputation to sell the idea of reliable AI answers. 

Even with the most powerful AI model at your fingertips, accuracy isn’t a guarantee. We’ve recently published a review on what it takes to get reliable analytics when Claude is connected to a warehouse. Without the right context engineering, accuracy lands roughly around 21%. 

Claude is capable. However, the context that feeds it determines whether that capability translates into trustworthy answers. And the context layer Snowflake is shipping only sees what's inside Snowflake.

What Databricks is betting on: Building data architecture for AI

Databricks Summit started with a different thesis: Enterprise AI fails not because of governance gaps, but because of data pipeline and architecture gaps. You cannot govern what you cannot move fast enough to use. Here’s how they plan to solve it:

Expanding Unity AI Catalog

Unity AI Catalog extends Databricks' data governance layer into the AI stack itself. Agents, models, MCP services, and skills now sit inside the same control plane that already governs your data. Cost controls, runtime policies, and end-to-end tracing come with it.

This is an undeniable win for AI observability. You get better controls to monitor and constrain agent behavior. What it doesn't change is the economics. Every agent query still consumes input and output tokens, and agentic systems run more queries than any dashboard ever will. Better visibility into spend is not the same as lower spend.

Introducing LTAP (Lake Transactional and Analytical Processing)

LTAP is arguably the most technically ambitious announcement of the summit, and the one carrying the biggest "if."

Databricks says this new data architecture unifies transactions, analytics, streaming, and operational data on a single copy of storage in the lake. One copy of data serves both real-time operational systems and analytical workloads. No ETL pipelines. No synchronization overhead.

If it works, it collapses one of the longest-standing trade-offs in enterprise data architecture. HTAP architectures have been promising the same outcome for years though, and most have never survived contact with production systems. LTAP's credibility will come from deployment evidence, not demos.

Adding Ontology to the Genie family

Databricks is taking a different approach to the semantic layer than Snowflake. Instead of manually governed definitions sitting in a catalog, Databricks added Ontology to its Genie Suite, which builds a living knowledge graph of how your business works. 

Ontology extracts structure from your data and exposes it to agents as operational context. The bet is that curated definitions can't keep up with how fast business logic changes, so the layer has to evolve. 

It's a sharper move than Snowflake's, and a more realistic view of how enterprise data behaves. However, the boundary problem still remains. Ontology only sees the structure inside Databricks. The relationships that matter most to your business often live in different systems.

Preparing your data strategy for AI in 2026 and beyond

Both summits made it clear: 2026 is the year of AI agents. Snowflake and Databricks are building governance, context, and AI infrastructure directly on top of their data platforms, betting that consolidation is the fastest path to enterprise AI.

As a CDO, the choices you make in the next 6 to 12 months will shape how much of your AI spend turns into ROI. Here are the top priorities you should anchor your AI data strategy around:

Budget for AI workload volume

License fees are the smallest line item on your bill. What drives spend is token consumption, query volume, model routing, and inference compute. Most teams don't discover these hidden costs until the first invoice arrives. Build your cost model around usage, not tooling. Treat cost controls as a visibility mechanism, not a savings one: they'll tell you where the money is going, but they won't slow it down.

Audit your team's capacity

Setup is the easy part. The harder question is whether your data team can maintain these systems a year from now, and how much re-skilling they'll need to do it. Every context layer, semantic model, and governance framework comes with its own conventions, tooling, and learning curve. The vendors assume your team will absorb all of it. Before you sign, audit your team's bandwidth, current skill set, and appetite for ramp time, and price those into the decision.

Looking to upskill on AI context? Register for our AI Context Engineer certification.

Plan for cross-system context

Each vendor governs the data inside their own walls. Neither one reaches Salesforce, dbt, or the other warehouses and tools an enterprise uses. If your AI data strategy spans more than one platform (and it almost certainly does), you need an Adaptive Context Engine that sits above the warehouse, not inside it.

Account for context decay

Metrics get renamed, ownership changes hands, new product lines introduce new vocabulary, and acquisitions add entire data estates with their own conventions. A semantic layer that was accurate six months ago will quietly lead agents to incorrect answers. That is, unless someone and something is actively managing your context to keep it current. Ask vendors how their context layer detects and resolves decay, not just how it captures definitions on day one.

Build accurate Agentic Analytics with an Adaptive Context Engine

The launch of Sense, Horizon, Ontology, and Genie all make the same case: an AI Analyst without context is a chatbot. With the right context, it's a data analyst who knows your business.

On this point, we agree. The question is, how do you deliver context from your entire data landscape? Without that, your AI answers are incomplete and therefore inaccurate by design, no matter how capable your model is. 

WisdomAI provides a simple way to ask questions and receive reliable answers from your data — not just the data that lives in Snowflake or Databricks, but across every other data source your business depends on. Our Adaptive Context Engine automatically builds, learns, and governs domain context so agents stay accurate, trustworthy, and explainable over time. 

See how WisdomAI customers deliver over 95% accuracy across their full data stack. Book a demo today.