Enterprise AI architecture is beginning to change in a significant way.
Until recently, organizations wanting to build generative AI applications typically assembled a separate technology stack around their existing data platform. Enterprise data would be extracted, transformed, moved into specialist infrastructure, connected to foundation models, and then exposed through applications built and governed elsewhere.
That architecture enabled experimentation, but it also introduced complexity. Every additional data movement created another security boundary, another integration to maintain, another source of latency, and another place where governance could become fragmented.
Platforms such as Snowflake are now challenging that model by bringing AI capabilities directly to the data.
Snowflake Cortex is a clear example of this shift. Rather than treating AI as a separate workload that sits outside the enterprise data platform, Cortex makes large language models, AI functions, document intelligence, semantic search, and agentic capabilities available within the Snowflake environment itself. In November 2025, Snowflake made a wider set of Cortex AI Functions generally available within its SQL engine, enabling organizations to process text, images, audio, and video without moving those workloads into separate AI services.
For Snowflake partners, Cortex signals something bigger than another product expansion. It points towards a future where analytics, data engineering, machine learning, and generative AI increasingly converge around the same governed data foundation.
AI is moving closer to enterprise data
The logic behind native AI capabilities is straightforward: enterprise AI becomes considerably easier to operationalize when organizations can bring computation to their data rather than repeatedly moving data to external applications.
Snowflake Cortex AI Functions allow teams to perform activities such as classification, extraction, summarization, sentiment analysis, translation, embedding generation, and similarity analysis using SQL and Python within Snowflake. Multimodal capabilities now extend that model beyond traditional structured data, allowing documents, images, audio, and video to become part of the same analytical environment.
For data teams, this has important architectural implications.
A customer service dataset no longer needs to consist solely of structured case records. Engineers can combine those records with transcripts, documents, emails, or images and apply AI functions within the same platform. A financial services organization might extract information from documents before combining it with transactional data. A manufacturer could bring structured operational records together with technical documentation or unstructured maintenance information.
This reduces the need to build bespoke pipelines for every AI use case and, critically, allows organizations to maintain more consistent governance around the data and models involved.
Snowflake’s current documentation states that the LLMs it makes available through its AI features are deployed within the Snowflake service perimeter. For enterprise organizations concerned about sensitive customer, commercial, or operational information, keeping AI processing closer to governed data can simplify some of the security and architectural decisions that previously made production AI difficult.
The principle is not that every AI workload must live inside the data platform. Rather, it is that the data platform is becoming an increasingly logical control point for enterprise AI.
Analytics, data engineering, and AI workloads are converging
For years, organizations often treated analytics, data engineering, data science, and application development as relatively distinct disciplines.
Data Engineers built pipelines. Analysts queried curated datasets. Data Scientists moved selected information into model development environments. Application teams then operationalized outputs elsewhere.
AI is beginning to erode those boundaries.
Cortex AI Functions are deliberately designed to resemble familiar analytical primitives. Snowflake’s AI_FILTER capability, for example, allows natural-language evaluation to be applied within SELECT, WHERE, or JOIN logic, while AI_AGG can derive insights across groups of unstructured records. AI_EMBED can generate embeddings for similarity search, and document-processing functions can turn unstructured files into information suitable for analytics or retrieval-augmented generation.
The result is that AI operations increasingly become part of a broader data pipeline rather than a separate downstream activity.
That changes what modern data engineering looks like.
Engineers still need to manage ingestion, transformations, orchestration, testing, modeling, governance, and performance, but increasingly those pipelines also need to prepare unstructured information, maintain embeddings, support semantic layers, manage AI-related access controls, and create reliable context for models and agents.
The same convergence is happening at the application layer. Snowflake has expanded beyond individual Cortex functions into Cortex Agents and Snowflake Intelligence, enabling AI systems to reason across structured and unstructured enterprise information and perform multi-step tasks.
In November 2025, Snowflake reported that more than 1,000 customers had already used Snowflake Intelligence to deploy more than 12,000 AI agents.
That is an important indicator of where enterprise data platforms are heading. They are no longer simply places where organizations store and analyze information. Increasingly, they are becoming environments where AI can interpret that information and help initiate action from it.
Bringing AI to the data does not remove the hard problems
The architectural benefits are compelling, but native AI does not make enterprise AI simple.
Moving models closer to data can reduce integration complexity, but organizations still need to solve questions around data quality, semantics, governance, cost, observability, and business context.
Data readiness remains particularly important.
Snowflake’s 2025 research with Enterprise Strategy Group found that 92% of surveyed early AI adopters said their AI investments were already producing returns, while 98% planned to increase AI investment during 2025. Yet 58% still said making their data AI-ready remained a challenge.
That tension captures one of the defining issues facing enterprise AI. Organizations can see the value, but scaling that value requires considerably more than model access.
AI systems need reliable context. If customer data is duplicated, documents are outdated, permissions are inconsistent, or business definitions vary between teams, moving AI capabilities inside the data platform does not make those issues disappear. In some cases, AI can amplify them because incorrect information can be interpreted and distributed at far greater speed.
Gartner identified poor data quality as one of the most frequently cited barriers preventing advanced analytics and AI deployment through 2025.
Partners therefore need to approach Cortex projects as data architecture initiatives as much as AI initiatives.
That means understanding which datasets an AI workload should access, how structured and unstructured information relates, where semantic context comes from, which users or agents should be permitted to retrieve sensitive information, and how outputs can be traced back to source data.
It also requires strong cost management. AI workloads introduce different consumption patterns from traditional analytics, particularly where high-volume inference, document processing, embeddings, or agentic workflows are involved. Teams need visibility into how use cases consume resources and whether the resulting business value justifies that consumption.
The modern data platform is becoming AI infrastructure
Taken together, these changes suggest that the definition of a modern data platform is expanding.
Data storage and analytics remain fundamental, but organizations increasingly expect platforms to provide a governed foundation for AI development, model access, semantic understanding, unstructured data processing, and agentic applications as well.
Snowflake’s product direction reflects that shift. Cortex AI Functions make model capabilities accessible through SQL and Python. Cortex Search supports retrieval across enterprise information. Cortex Agents can orchestrate complex tasks across structured and unstructured data. Snowflake Intelligence provides a natural-language interface through which business users can interact with that information.
Snowflake’s FY2026 reported more than 9,100 accounts using its AI features, with Snowflake Intelligence reaching almost 2,500 accounts within three months.
For partners, this creates significant new opportunities, but it also means Snowflake projects are becoming more multidisciplinary.
A successful implementation increasingly requires teams capable of understanding data engineering, governance, AI-ready data design, model interaction, retrieval patterns, security, and business process requirements together. Platform expertise alone remains important, but the ability to apply that expertise to AI workloads is becoming increasingly valuable.
What this means for Snowflake partners
The most immediate implication for partners is that customer demand is likely to shift from isolated AI experiments towards integrated data and AI programs.
Customers may initially approach a partner because they want to implement Cortex, build an AI application, or explore agentic use cases, but the underlying work frequently reaches much further into the data estate.
- Is the source data reliable enough?
- Are pipelines designed for the latency the AI use case requires?
- Can structured and unstructured information be connected effectively?
- Are access controls appropriate for AI workloads?
- Does the organization have the engineering practices required to move an application from proof of concept into production?
These questions place Data Engineers at the centre of AI delivery.
Rather than becoming less important as platforms automate more functionality, strong engineering capability becomes more valuable because organizations need people who can connect new AI capabilities to production-grade data foundations.
The skill profile is evolving accordingly. Snowflake Data Engineers increasingly benefit from understanding not only ingestion, modeling, SQL, orchestration, and governance, but also machine learning foundations, generative AI-ready data practices, unstructured data, and the architectural implications of deploying AI against enterprise information.
For partners trying to scale that capability, relying exclusively on experienced external hires can become difficult as demand expands faster than established talent pools.
Building Snowflake talent for AI-era delivery
Revolent helps Snowflake partners address this challenge by developing certified Data Engineers with the practical skills required for modern Snowflake environments.
Our Snowflake talent program was built in collaboration with Snowflake and combines Snowflake and dbt certification with practical capability across data ingestion, orchestration, governance, modeling, machine learning foundations, and generative AI-ready data practices. Training also covers technologies and techniques such as Airflow, Airbyte, FiveTran, scripting, and visualization, helping professionals understand the wider data engineering environment around Snowflake.
Crucially, the program is designed around practical delivery rather than certification alone. More than 50% of training time is spent on use-case-driven practical application, and professionals continue developing after deployment so their skills can evolve alongside Snowflake’s platform and the AI projects they are supporting.
That ongoing development matters in an ecosystem moving as quickly as Snowflake. The skills required to deliver a warehouse modernization program several years ago are not identical to those needed to support Cortex, generative AI-ready data environments, machine learning workloads, and increasingly agentic applications today.
Partners therefore need talent strategies that evolve alongside the platform itself.
Enterprise AI is becoming a data platform problem
Snowflake Cortex provides a useful indication of where enterprise AI infrastructure is heading.
Models will continue improving, and organizations will continue experimenting with new AI applications, but the long-term challenge is increasingly about how those capabilities connect to enterprise data securely, efficiently, and at scale.
By bringing AI functions, model access, unstructured processing, search, and agents closer to governed enterprise data, Snowflake is reducing the distance between analytics and AI. In doing so, it is also changing the skills required to build and operate modern data platforms.
For Snowflake partners, the opportunity is not simply to help customers switch on new AI functionality. It is to help them build the data architecture, engineering capability, and governance required to turn those capabilities into reliable production systems.
And as AI becomes a native part of the Snowflake platform, the partners that build those capabilities now will be better positioned to deliver the next generation of enterprise data and AI projects.
To learn more about how Revolent can help you build Snowflake Data Engineering capability for modern data, machine learning, and generative AI projects, get in touch.
