Snowflake AI: Snowflake Cowork & Custom Skills

[00:00:00]Hello everyone. Welcome to this video of mine. So today in this video I'm going to talk about Snowflake's new announcement which is Snowflake Cortex.

[00:00:12]And within Snowflake Cortex I'm going to show how can we configure custom skills within this console.

[00:00:21]So what is Snowflake Cortex? So formally it was known as Snowflake Intelligence and only recently in the summit announcement it has been rebranded as Snowflake Cortex. And again it is an extremely powerful agentic capability within Snowflake as a platform which can plan, which can you know track your details. It has got you know kind of ability to you know kind of take lot of tools and get you know itself integrated. So all these capabilities can be possible within Snowflake Cortex, okay? And then you know if you see this there are certain key features that it has. So just for the demo sake today I am going to talk about different skills which we can definitely use in Snowflake Cortex and how it can be configured, okay? So without further delay let's quickly go to the demo and see how we can configure the skills within Snowflake Cortex, okay?

[00:01:26]So this is where I am right now in Snowflake console and if I go to AML section I will see two options agents and Snowflake Cortex. And this one that you see Snowflake Cortex it was earlier you know kind of termed as Snowflake Intelligence but now it is Snowflake Cortex, okay? Now if I go to agents, okay, what you know I will have let's see, right?

[00:01:55]So there are two agents which are configured over here. Okay? Now, the for the demo sake, I am going to kind of do all the configuration in this particular agent. Okay? Now, the agent name is agent sales agent {underscore} sales {underscore} analytics, right? Now, how this one is related to Snowflake Co-Work? Now, once we create any agent within Snowflake, so what will happen is we can you know, kind of integrate that particular agent in Snowflake Co-Work.

[00:02:30]So, this is the option. If you see this, Snowflake Co-Work is there and it is available. So, what we, you know, usually do is once we configure this agent, right? And then easily we can ensure that this agent is only looking up to, you know, kind of only, you know, associated with the Snowflake Co-Work as a service, right? So, this is exactly how, you know, the mapping to the Snowflake Co-Work and agent is done, right? Now, over here, what I will do is I have I'll I'll I have edited the section of this agent. And then I'm going to the skills section. Now, over here, what we see is nothing is added. None of the skills are added over here, right? So, how do I add skills in the agent is something that we'll see, right? And it is not that without skill this particular, you know, kind of Co-Work Snowflake Co-Work or agent will not work. It would work. But, when we start adding skills, what it will give is additional insights, you know, kind of purely curated insights.

[00:03:41]What exactly do we need from the data sets that we have, everything will be given to us, right? Now, for me, this data set is primarily a sales data set on I have defined this agent. That means what? So, think of it like uh in a region, if I am having lot of sales, lot of deals, lot of registrations, how many are won, how many are lost, what are the reason reasons behind it, what is the percentage of wins, everything is there under the data set, and on top of it, I have defined this agent, okay? Now, over here, what I want to do is I want to add one custom skill.

[00:04:21]So, how do you do that? You have to edit it, and then go to this console, and then add a skill, okay? And there are three options through which we can add.

[00:04:30]Directly, we upload it from the computer, add it from the stage, or we add it from a Git repository, okay? Now, what I will do is I will add it from the stage. And now, when I add from the stage, okay?

[00:04:44]So, this will take me to a console wherein I obviously have need to have the database and schema, which is intelligence demo uh is my database, sales {underscore} analytics is my schema, and all my folder and my file has to be present in that internal stage, okay?

[00:05:04]Now, if I see this over here, I will have a stage called as agent skills, right? And then if I go to skill folder, I will have a you know, kind of folder name deal {underscore} win {underscore} loss {underscore} analyzer, right? And then, that is exactly how I do the additions in here. Now, let us see what is present in this internal stage, which is agent skills, right? So, for that, let me go to the same you know, kind of account, but I have opened a different session.

[00:05:40]So, what I will do is everything was present inside intelligence demo, okay?

[00:05:46]And then, what I do is I go to sales analytics.

[00:05:49]And then what I go to I go to stages.

[00:05:52]And then under stages I have agent {underscore} skills, right? Now, this is what it is, right? Deal {underscore} win {underscore} loss {underscore} analyzer. Now, this is not a file. This is a folder, okay? Under this folder, I am having skill.md.

[00:06:11]Please ensure that whenever we are uploading any, you know, kind of custom skill it has to be represented in this format, because skill.md is nothing but the markdown file. Snowflake understands the custom skill with respect to markdown files. And then this is the skill name. Deal {underscore} win {underscore} loss {underscore} analyzer is a skill name. So, that is how you know, Snowflake will understand that this particular skill has been uploaded into the system. And whenever I will start using Snowflake co-work, it will look up to this particular skill name, and then this markdown file, under which I have written all the custom instructions, okay? Now, I can even, you know, show, you know, uploading one particular at least you know, kind of file. So, let's do that, okay?

[00:07:07]So, what I will do is let me add a, you know, kind of custom file, right? Now, over here, if you see this uh it will ask me directly to upload something, right? I have to browse certain things and I have to upload.

[00:07:21]That means what? I have to upload my files. If I directly upload any file in this, you know, kind of agent skill, I will not be able to read that skill.md.

[00:07:32]That is why I have to create a subfolder, okay? Now, let's say I will create something like demo {underscore} test {underscore} 001, okay? And then I will upload the markdown file, which is nothing but your exact content in which I have defined the custom skill code, okay? Now, if I do this, then what will happen is the folder will be there. Along with the folder, the markdown file will be also be there, okay? Uh this becomes your, you know, kind of custom skill, which is demo test _001, okay? So, that is exactly how you upload, you know, the custom skill through the internal stage, okay? Now, this exactly can be done if I am uploading anything or if you're uploading anything from our local machine and through, you know, Git. Uh but only thing is, please ensure that the custom skill that we create, it needs to have one folder name and under that folder, we need to have the skill.md, which is the, you know, kind of core of custom skill, right? Now, I've done. So, this particular, you know, kind of agent will now be enhanced with this custom skill, and the custom skill name is deal_win_loss_analyzer, right? And I will add it now. That means what?

[00:08:57]My skill has now been added and it has been added from the internal stage, okay? Right? Now, let me kind of go to the tools and see whether everything is there. Let me go to the about section.

[00:09:11]Let me go to the skill section.

[00:09:12]Everything is there. MCP connectors, orchestration, right? Everything is there for now, right? So, let me kind of save it, okay?

[00:09:21]So, once it is saved, I will also show you the, you know, kind of full configuration of this agent, okay? Now, it is saved right now, okay? Now, Snowflake co-work is available, right?

[00:09:33]So, let me see, right?

[00:09:36]Fine. So, this display name is agent_sales_analytics and it is now, you know, available through co-work, right? Now, let me kind of see whether my skills, you see this?

[00:09:49]My skills also got added over here, okay? And I don't want to kind of do any, you know, kind of interactions uh of uh you know, this particular agent through here. Rather, I will kind of do the interactions through my Snowflake co-work, okay? Now, if you see this, if I open Snowflake co-work, okay?

[00:10:12]What will happen is I will get a new console, okay? Now, in this new console, the same agent is there, okay? Agent_sales_analytics, for which we have just now done all the configuration. And this is what my new session has been opened, okay? Now, let me let us, you know, kind of see how the skill with and without skill, the same prompt gives us the output, okay? Now, uh the prompt that I'm using right now, "Show me the count of orders by region and by deal stage, okay?" So, there are various stages of the deals, right? So, what it will try to do is it will give me a count of orders by region and by deal stage. And what is this? Snowflake co-work, right? This is a powerful conversational interface through which I am able to access the agent and ask the questions in natural language, which implicitly looking up to that agent and that agent is talking to the different data sets in which I am having all the insights that this particular prompt is going to give me, right? So, the entire, you know, think of it like the query formation, the planning, the instructions, everything will be done implicitly by Snowflake to work for us, right? Now, if you see if let let's see the output of this one, right? Show me the count of orders by region and deal stage. So, it gave me output, right? So, region is there, right? In this region, how many deals we lost? North America was the region. How many deals we won?

[00:11:52]Total.

[00:11:54]And again, this metrics has been split across different region, right? And again, uh what it also gives is you know, kind of some of the key observations like Europe has the highest number of order volume. Latin America has the fewest number of orders. All these things are fine, right? But, you know, it is not able to give me some metrics with respect to percentage split, right? Some metrics with respect to at least pictorial depiction of how the you know, kind of region to deal stage has varied, okay? That is exactly where this you know, kind of custom skill will come into picture. I'm talking about custom skill, not custom tool, okay? Now, how to access custom skill? You just have to you know, kind of type forward slash and because we have integrated this particular deal win loss analyzer, I'll have to just click it. And what I did was I have typed the same prompt, but now I am asking this prompt to use this particular custom skill also, right? Let's see, you know, what is the output that it you know, kind of gives us, right? So, what it will try to do is it will try to give me the same insights, but now it will also use this custom skill and let's see whether there will be any difference because the first one, obviously, it was still powerful. It was still giving me the response that I wanted.

[00:13:24]But, with custom skill, I am trying to get very very, you know, kind of curated output. Your focused output. What exactly, you know, and again, custom skills are custom skills because I can always play around with the custom skill. How do I want to, you know, kind of show the deal win, whether it will be in percentage form, whether it will be in graphical representation form, so on and so forth. You see this?

[00:13:50]In this custom skill, you know, I have written a, you know, kind of prompt that I need everything in pictorial representation. That is exactly where it gave me a bar chart view of how the order count by region and deal stage varies, right? And then also it is give me it is giving me the summary by region with respect to win rate percentage, okay? So, this becomes an impactful, you know, kind of metrics with respect to visualization and with respect to win rate percentage also. That again will help me take informed decisions, right? So, that is exactly how powerful Snowflake Co-work is. And when we start using custom skill, you know, uh you through it, I am definitely going to get very very powerful, you know, kind of metrics and it customized metrics, right? I will be able to at least, you know, kind of channelize my prompt and do the instructions and give the instructions in a way that I get the, you know, curated insights, which is exactly needed for my business purpose, right? And all these things are very very configurable, very easily configurable. Uh so, today what I focused was how do you configure skills in Snowflake Co-work. And again, right? If there are any doubts, you know, with respect to configuration, we anyways have Snowflake Kuku wherein we can definitely ask lot of questions, right? Over here, if you see this, I have I I keep on asking lot of questions, right? Example, if I want to kind of configure even some of the skills, so what can be the basic steps through which, you know, I can do those configurations? Snowflake uh Kuku can definitely help us with respect to that understanding, right? So, again, so this summarize, this video was more about to make everyone understand how powerful this platform has become with respect to agentic AI capabilities after, you know, kind of submit announcements and how easy it has made uh this platform features so that anyone and everyone can start using Snowflake agentic uh you know, kind of features for solving their various business use cases.

[00:16:20]So, thank you so much once again for watching this video and staying with me.

[00:16:25]If you like this uh video, please feel free to like, comment, share, and subscribe to my channel. Thank you so much once again.