Snowflake sees itself as the ideal partner to manage the rollout of AI at an enterprise level. Context, governance, and security are essential for this, so the company wants to provide the tools for it. However, this is a story we’ve heard before, so what makes Snowflake the right fit?
On stage at the Snowflake Summit in San Francisco, CEO Sridhar Ramaswamy explains what an enterprise needs to fully embrace agents. Approximately 20,000 attendees listen to his analysis. According to Ramaswamy, three components are indispensable: data with associated context in an enterprise-grade platform, AI models, and software and applications.
Nobody wants walls
That is not a groundbreaking insight. We hear that high-quality and accessible data is essential from the likes of Salesforce, ServiceNow, or SAP. Almost all major players have now torn down the walls around their data and are investing in zero-copy integrations with each other’s systems. Those who have their data strategy in order can link systems and get them ready for AI.
It goes without saying that AI models are important. Ramaswamy points out that being able to switch quickly between models is an asset. Snowflake is going all-in on an LLM-agnostic approach. Daniela Amodei may take the stage a bit later, but Anthropic is not an exclusive partner. OpenAI models are also available, as are open-source alternatives, LLMs from the French company Mistral, and as of this Summit, even AI models from SpaceX.
Context over model
More important than the model used is the context in which the model is used. Thus, the second pillar is linked to the first. At the Summit, Snowflake is paying a lot of attention to the introduction of systems that provide extra context to AI agents and assistants.
This specifically concerns Horizon Context and Cortex Sense. These systems look at data inside and outside Snowflake and can provide queries and prompts with relevant business context. Christian Kleinerman, EVP of Product for Snowflake, points out that this significantly increases the accuracy of a prompt.
The relevance of context for AI accuracy is a concept the entire industry is starting to agree on. A few weeks ago, we heard the same sentiment at SAP: a knowledge graph (like Horizon Context) with knowledge about the company takes the performance of an AI system to a higher level. AI agents and assistants have to guess less and interpret prompts in a more targeted way.
Connection with software
Software and applications must also be linked to the AI story. Data and insights from your email applications and external systems are relevant, which implies that tools need to be linked via a compatible platform. Nothing and no one works in isolation anymore: SAP cited its connection with Snowflake to prove this, and Snowflake is doing the reverse. Integrations are numerous: the mutual connection with Salesforce is another example.
This creates several possible centers of gravity. LLM builders seem indispensable but interchangeable in the context of enterprise AI. However, Snowflake, SAP, Salesforce, and ServiceNow are all telling a similar story. They all want to make a difference, and they are all doing it in the same way.
Who gets to be the glue?
Like other organizations riding the AI hype, Ramaswamy has realized that a fourth component is needed for enterprise AI to flourish. Data and context, models, and applications need to be connected by a control layer that links agents, allows for management, and provides the control mechanisms to keep them in check.
For Snowflake, CoCo (formerly Cortex Code) and CoWork (Snowflake Intelligence) are the glue. Oversight comes from the Trust Center. Furthermore, Snowflake has further developed its platform architecture tailored to the management of agents. The introduction of Agent Identity is very relevant in this regard: it allows the Snowflake platform to treat agents as a unique entity, so that organizations can link custom policies to them.
Snowflake tells the story in its own way, but we hear echoes of SAP’s AI Agent Hub and ServiceNow’s AI Control Tower. The role of CoCo and CoWork is not entirely comparable to that of Joule at SAP, but in both cases, it involves an AI-driven layer that connects everything.
Claiming the same central position
It is striking that organizations are gradually agreeing on what is needed for the rollout of secure, scalable, and accurate AI at an enterprise level. Now that the puzzle is coming together, every major player is trying to claim from their side that they are uniquely positioned to be the central hub in the whole story.
We have an advantageous position thanks to the knowledge we have about our customers.
Sridhar Ramaswamy, CEO Snowflake
“We have an advantageous position thanks to the knowledge we have about our customers”: Ramaswamy says it at the Summit, but we’ve already heard it at ServiceNow, Salesforce, and SAP.
A touch of realism
So, who holds the best cards? Realism suits the CEO of Snowflake. “Both application companies and model builders can claim the crown,” he admits. Snowflake argues that it has the ideal starting position through its data platform. “In a few years, it will be clear how the market has evolved, and then we will all find it obvious, but we have to wait and see.”
That doesn’t mean Snowflake can’t support its claim to the lead with several arguments. Co-founder Benoît Dageville expresses this best. “The original goal of Snowflake was always to break down silos, as well as the causes for the creation of those silos,” he outlines.
Snowflake has already done this several times. The platform makes data sharing secure and simple, both internally and externally. With Apache Iceberg, Snowflake embraces openness and interoperability. Furthermore, the platform runs on the infrastructure of the three major hyperscalers and is thus cloud-agnostic, so that no silos are created there either.
The role of the data platform
According to Dageville, one thing is crucial for AI: “Data and AI must live in the same place.” In other words, the computing power for AI must be next to the data and linked to the data platform. “Otherwise, we create complexity again, and silos will emerge once more.”
If that is entirely correct, Snowflake indeed holds a strong hand. In that context, the data platform can become the foundation of AI development, with the data and applications of other players being linked to Snowflake. The Trust Center, CoCo, CoWork, Horizon Context, and all other innovations will support the rollout of AI in that case.
Everyone is right
Presumably, everyone is a little bit right. Organizations that are deeply embedded in the Salesforce ecosystem will drive innovation from there. In that case, it seems logical that the Salesforce Agent Builder and the Einstein Trust Layer will take the lead. Manufacturing companies with a large investment in S/4HANA will be more likely to believe SAP when it claims to know its customers inside out. In that case, Joule will likely win out over Cortex, CoCo, and CoWork.
For organizations that view their digitalization and AI projects from a data perspective, it makes sense to start there. In that case, Snowflake can position itself as the central hub.
No lack of enthusiasm
Furthermore, it seems that Snowflake customers are enthusiastic about it. Just over 13,600 customers use Cortex AI weekly, according to Timo Meijrink, Head of Solution Engineering in the Benelux. Out of a customer base of approximately 14,000 organizations, this shows both enthusiasm and a low barrier to entry.
As Ramaswamy indicates, we won’t know immediately who will emerge as the central hub in the AI transformation of organizations. That Snowflake can claim the position is beyond doubt.