Snowflake’s head of AI takes an inside-out approach to developing AI solutions

Snowflake is first and foremost a data platform, a place where companies can store, analyze and share data in the cloud. That puts the company in the right place at the right time as data has become central to building meaningful solutions on top of large language models, whether that involves a ChatGPT-like interface or agents. 

And if AI is the most significant tech trend right now, being head of AI at a successful cloud data lake company like Snowflake puts Baris Gultekin in the middle of this massive shift. He joined the company when his startup was acquired in 2023. He recognizes how fortunate the timing was, coming right after OpenAI launched ChatGPT and changed the way we think about software forever.

While it’s an important job building AI applications for customers, Gultekin also needs to think about how the company can apply AI internally to run the business. He sees Snowflake itself as customer zero, giving the company crucial first-hand insight into how AI can impact a business user. 

“AI is moving incredibly fast, and one of the things we pride ourselves in is that we make AI easy for our internal teams,” Gultekin told FastForward. These teams tend to work independently building AI tools, and his team meets regularly with other departments to understand and help refine these projects. 

In this way, Gultetkin sees his role in a similar way as Salesforce CIO Juan Perez, whom we interviewed last year for FastForward. “I have a dual responsibility at Salesforce: one is that I have to push for Salesforce and the Salesforce platform, and I have to make sure that I take my responsibility as ‘customer zero’ very seriously,” Perez told us.

The trick is to take some of those internal deliverables, and what the teams have learned creating them, and put them into practice for customers, as well.

Customer support

An important aspect of being head of AI at Snowflake involves communicating with the product teams, and making sure he’s aware of everything going on in the company involving AI. “I oversee our product efforts including both generative AI and predictive models,” he said. “It is such a broadly applicable technology with a broad spectrum of use cases.” So his team tries to develop products that could be useful across a variety of business scenarios.

That involves building their own models, as well as using partner models to give their customers the widest choice possible. The goal is building an end-to-end system to give their customers the ability to choose the right model for the job.

“We have a series of models that we build ourselves. For instance, we have a search model that's doing retrieval – it's top of its class from a retrieval perspective – and we got that through the Neeva acquisition,” he said. 

Yet they also recognize customers want to work with models from other sources, depending on their requirements, and that’s where the partnerships come in. For instance, the company made a high profile partnership announcement with Anthropic in November, and more recently announced a partnership with Microsoft to access OpenAI on Azure. (Microsoft has a deep partnership with OpenAI.)

Bringing agents to the platform

As companies move from pure generative AI projects, simply asking questions about the data in plain language, into agentic AI that can take actions on that data, Snowflake wants to be involved there too. Every major company it seems is making a big agentic AI push right now. We recently saw agent-related announcements from Salesforce, Microsoft and AWS.

While there isn’t a standard industry definition for agents, for Snowflake it involves autonomous and intelligent software that can perform a series of tasks with little or no human intervention. “We think of agents as these models that are able to reason and take action and have a state, have memory. So when you bring all of this together, these agents are very capable,” he said.

Regardless of how customers are implementing agents, Gultetkin recognizes we are still in the very early stages of the development of this technology, and they are building a platform where customers can create their own agents, or use prepackaged offerings, depending on the requirements. 

Snowflake offers API access to data stored on the platform to make it easier for customers to build AI solutions themselves. More recently, the company announced Cortex Agents as part of their Cortex AI platform, which the company describes as enabling “AI developers to build AI data agents with code.”

We think of agents as these models that are able to reason and take action and have a state, have memory.

But some customers don’t want to do all that work, especially those who aren’t developers by trade, so Snowflake created a product aimed specifically at business users. “More and more we're seeing our customers want packaged, ready-to-go tools for their business needs,” he said. As an example, in November the company announced a new tool in preview called Snowflake Intelligence, which is aimed at less technical users.. 

“This is a product we're building that allows our business users to point to their data in documents inside Snowflake, and then all of that data becomes available to be used with an agent, which becomes an assistant for their business users,” he said.

Regardless of the use cases, or whether the product comes from inside the organization or from customers, Gultekin needs to keep his finger on the pulse and understand how everyone is putting AI technology to work, while continuing to monitor the fast-changing AI landscape to come up with solutions that help both groups.

Feature image of Baris Gultekin courtesy of Snowflake.