Snowflake Summit - why founder Benoît Dageville was disappointed by the rise of ChatGPT

Mark Samuels Profile picture for user Mark Samuels June 29, 2023
Snowflake co-founder says the tech company wants to help its users exploit AI.


Snowflake is refining its cloud-based platform to give enterprise users the opportunity to take advantage of advances in Artificial Intelligence (AI).

Benoît Dageville, co-Founder and President of Products at Snowflake, explained his vision to diginomica in an exclusive one-to-one interview at Snowflake Summit 2023 in Las Vegas. Dageville covered a wide range of topics, including the growing importance of AI, the developments of Snowflake’s app store and the ever-growing importance of containers.

One of the big headlines from Snowflake CEO Frank Slootman’s keynote speech was the company’s move into AI and its use of large language models (LLMs). Snowflake purchased generative AI specialist Applica in September 2022. The tech giant is using Applica’s multimodal LLM in its new products, including Document AI

While it’s currently in private preview, Dageville says Document AI will help customers find insight in unstructured data. And while every tech company seems to be making similar moves into emerging technology right now, Dageville says Snowflake’s move into AI is not just a case of following the herd:

People are trying to react – and we are not reacting. We didn't discover AI when ChatGPT exploded. We had been working on it for quite a long time. And actually we were disappointed by the rapid rise of ChatGPT, because if we’d presented Document AI earlier, everyone would have said, ‘Wow, this is amazing. You can query natural language and complex documents with writing and images. This is amazing; I’ve never seen something like that.’

Dageville stresses that Snowflake’s interest in AI is not just confined to its own products, such as Document AI. He says the company’s moves into AI should be seen as a complement to its continuing attempts to give customers an all-in-one platform, known as the Snowflake Data Cloud, to store and exploit treasure troves of data:

Now, where we are focusing – and this has been the North Star of Snowflake from the beginning – is on bringing the workload to data. And obviously, AI is going to be one of the most critical workloads. At the same time, you cannot process these workloads if you don't have ways to transform data. You cannot build your AI if you don't have you a system where you can see all the data together.

That’s where Snowflake comes in, says Dageville, who suggests the company’s Data Cloud – with its set of services and functionalities – will be a strong foundation for the kinds of AI and machine learning-based developments that companies will build in the future:

I would imagine most apps – if not all apps – will be powered by AI. Now, we believe the way they are going to integrate AI deeply inside their applications is by using the building blocks that the Snowflake Data Cloud provides. Our users are going to have integration services, so they can do fine-tuning of their data and stitch complex machine-learning models together.

Building an app store and embracing containers

It was also announced during the keynote that the Snowflake Native Application Framework is now available for developers to build and test their own apps. Dageville discussed the early stage of these developments with diginomica at last year’s event. Today, he’s “super proud” of the progress that’s been made. He says Native Apps will help set the future direction of travel for Snowflake’s platform, leading to a “blur” between the capabilities the tech company provides to customers and the applications that third parties create:

It’s transforming Snowflake into something like an iPhone. I use this analogy because we are providing a platform – we are building the extensibility, so that people can build their apps and amazing data products.

Native Apps can run inside an enterprise user’s Snowflake account, which means customers no longer have to export or provide external access to data. Dageville says building this app store-like capability was challenging. However, 25 new Snowflake Native Apps are now available for customers and there are hundreds more on the horizon, so the preparatory work has laid the foundations for some big developments:

We stepped back and thought, ‘OK, what does it mean to run full apps in our Data Cloud? What would it take? What are the capabilities we need? What you need is the orchestration, the network, the routing and the security. Because if we are running many services in our Data Cloud, we must have the right level of security.

As well as a commitment to AI and apps, Snowflake also announced the launch of Snowpark Container Services, which expands the tech company’s compute infrastructure to run a variety of workloads, including full-stack applications and the hosting of LLMs. While the support for containers should be seen as an important step, Dageville says the ability to use a range of infrastructure options is more like a refinement of the long-term development efforts that Snowflake has worked on for the past decade:

In some ways, Snowpark is not a revolution because we didn't create another engine as a compiler. We wanted to leverage all our 10 years of investment – use all the optimization and the performance. We wanted to make it available to run Spark-like workloads and this is why, when people try Snowpark, they are amazed about how much faster and cheaper it is compared to whatever technology they used.

Dageville argues that the combination of developments – from AI to apps and onto containers – shows how the Snowflake stack continues to evolve in a holistic manner:

Now, we have everything – we have the data layer, which is Snowpark, then we have Snowpark Container Services, where you can host code, which is actually the middle tier of your application, or it's AI or it’s an LLM. It doesn't matter to us what’s running because, at the end of the day, it's inside this container.

My take

Bold ambitions. Dageville says the ongoing refinement of Snowflake’s Data Cloud and the features the company provides to its customers should go a long way towards silencing anyone who still think of Snowflake as being simply a ‘database company’:

I think people are going to realize – and people who are using Snowflake for a long time already know – that we are way beyond the database. Now, with Snowpark and Container Services, you can run fully fledged applications in our cloud. So, it might take some time for people who think we’re a database company, but they will get there.

A grey colored placeholder image