![[Presentation Materials] I presented at DevelopersIO 2026 Osaka Day 2 with the title "Understanding the Strengths of Snowflake Cortex AI by Building LLM Features from Scratch" #devio2026](https://devio2024-media.developers.io/image/upload/f_auto,q_auto,w_3840/f_auto,q_auto/v1791081720/eyecatch/dwyoltcoauwrcccvpvrk.png)
[Presentation Materials] I presented at DevelopersIO 2026 Osaka Day 2 with the title "Understanding the Strengths of Snowflake Cortex AI by Building LLM Features from Scratch" #devio2026
This page has been translated by machine translation. View original
This is Suzuki from the Data Business Division.
I gave a presentation titled "Understanding the Strengths of Snowflake Cortex AI by Building LLM Features from Scratch" at DevelopersIO 2026 Osaka Day 2, held on Tuesday, September 29, 2026.
Presentation Materials
Key Points
1. Strengths of Cortex AI
I introduced features that I have built myself and that represent unique strengths of Snowflake Cortex AI.
Cortex AI itself has been GA since June 2024, now about two and a half years ago, and has evolved rapidly by capturing user needs.
In particular, although it has been less than a year since Cortex Agents and Snowflake Intelligence (now CoWork) went GA last November, it has become possible to approach various business challenges with data and AI.
AI functions allow you to use high-performance AI capabilities from SQL and other interfaces. With this native feature, LLMs can be used very easily.
With Cortex Agents + Semantic View, anyone can easily use a Text2SQL mechanism to delegate data analysis to AI. Compared to when we were experimenting with Text2SQL back when LLMs first appeared, the progress is remarkable. Analytical SQL tends to be long and complex, but by using this mechanism, it is possible to generate SQL while taking business metadata and aggregation definitions into account.
Finally, CoWork is also extremely powerful.
The UI for using AI agents within an organization is very important, and there are many technical elements required.
You want to view conversation history in threads, and feedback functionality for agent responses is also necessary. Analysis results need to be displayed in graphs. Agents must provide analysis tailored to the permissions of the requesting user.
With CoWork, all of this is already in place, making it very easy for organizational members to conduct analysis and make decisions using agents.
2. Features I Am Looking Forward to in the Future
I introduced two features that I am looking forward to in the future.
The first is AppRuntime.
AppRuntime itself is a hosting feature for Next.js apps on Snowflake.
But beyond that, it benefits from authentication and governance features integrated with Snowflake, and users can deploy an app from implemented code with just a single command.
Also, particularly noteworthy is the implementation via CoCo. CoCo is equipped with snowflake-apps skills, making it possible to request app development through to deployment from AI via an interface like CoCo Desktop.
With features like these, while lightweight apps will likely be the main use case, anyone can now easily deploy apps on Snowflake.
The second is Cortex Sense.
This feature is still in the announcement stage, but it is described as a context layer that can pass data on Snowflake and uploaded business process information to CoCo, Cortex Agents, and other tools.
I often build Cortex Agents for analysis purposes, but currently, to pass business information to Cortex Agents, you need to configure things like tool usage order in Orchestration instructions. Once Cortex Sense arrives, business process information and more data from Snowflake will be available through Cortex Sense, and I look forward to Cortex Agents being able to better understand and act on user intent.
In Closing
That was an overview of the presentation content and key points from "Understanding the Strengths of Snowflake Cortex AI by Building LLM Features from Scratch."
I also introduced Snowflake Cortex AI last year, but at that time, many of the features introduced were still in public preview.
Currently, CoWork and Cortex Agents have also gone GA, and they can be increasingly used to solve challenges in real business situations.
In addition, powerful features like CoCo and AppRuntime have been added and gone GA, and the platform is evolving at a very rapid pace.
Let's make the most of these features and use Snowflake Cortex AI to solve more and more business challenges!
