Turn your LLM into a Customer Expert with MCP

Hey, everybody. How's everyone feeling? Anybody tired yet? Yeah, today I'll probably approach things a little bit more casually just because I'm tired.

It's been a long couple days. But I'm really excited about this session. I know MCP's been thrown around a lot. It feels like AI two years ago as a word and agents last year as a word.

That was just like all the time. That was the buzz word. And MCP is the buzz word right now. But it's because of all that it unlocks.

And so I am very excited about this topic because-- well, first off, how many of you in here are using an AI that's connected to MCPs? OK, good. That's good. Now, second off, how many of you in here are using an AI that's connected to either Gainsight or Staircases MCPs?

OK, sweet. This workshop's going to be a little bit different than a standard workshop because it is focused on use cases around the CS and Staircase MCPs. But there's a lot of general knowledge about MCPs and skills, is what they're called. And I'm going to go through a lot of education around that first before we go in.

And I'll model, but I'm also going to share some things that you all can download and use right now, especially if you have our MCPs that you have. So that's a little high level thing. Now, first, I want to give everyone value. And so this is either a skill or a prompt, like custom instructions for a project.

You can create a project in Chat GPT or in Cloud or anywhere else. So there's both a prompt or a skill around this. I will say this has kind of changed my life. I've never been one to cook other than like Hamburger Helper or other things like that, maybe making spaghetti or pancakes and sometimes all grilled to me.

And that's kind of been my role in the kitchen. Over the last few months, my brother, Garrett, introduced me to his-- he had it as a GPT, Chef Gordon. And essentially, this is a very abusive LLM that as you-- one, it'll give you really great recipes and walk you through it step by step. But it's also going to verbally abuse you as you do it.

So it will cuss at you. It will say really mean things. If you don't want that, you don't have to download this. But if you want to, it makes me chuckle every time that I'm cooking and in the kitchen.

And I'm cooking regularly. And honestly, I've made some really great food. I went to a potluck on Sunday. And I made lentils.

I've never made lentils in my life. And the lentils were delicious. They were the thing that people loved the most. It was-- it's really, really fun to cook, it turns out, and to make great food.

The thing I love most about it is that compared to like a recipe where it's like fixed and it's there, while I'm in the middle of cooking, if I'm like, ah, is this done, is this not, what do I do next, all those kinds of things, I can just ask. And it answers and it tells me the proportions or whatever else I need to do at that step. So have some fun with Chef Gordon. And I hope you enjoy it.

That also takes you to my GitHub repo that I just launched this week. It has also-- this is essentially a preview of our Gainsight plugin that I've been working on. Plugins in Claude allow you to essentially link to a repo. And in that repo, you can store skills.

If you're not aware of skills, I'm going to talk about skills. But skills are custom instructions, or it's kind of like a prompt, but it's a complex prompt with progressive disclosure and a few other things like that. And so I've been working really hard on integrating the Gainsight CS and Staircase MCP now that both are available, and the use cases around that. And I'm going to talk about more how I did it, but you can access all of it here.

It also has instructions on how to install the MCPs for-- if you have Gainsight and you're allowed to install MCPs. I know sometimes that takes security and governance, things like that. So feel free to take your time. So I've been a CSM before.

I've been a CS leader before. And I've worked in business for a long time now. And this has always been a problem, is needing to gather all the context for what I need to do to allow me to even do it. And things are scattered all over the place.

They're either in notes, manual notes, or notes that I've taken in my computer, or they're in Gainsight, or they're in another system in our CRM, or in CODA for us at Gainsight. They're all over the place in that way. And so you need to navigate to a bunch of different pages, even within an individual platform, you have to do that. In Gainsight, I go to the C360, but then I need to go to Reports and Dashboards.

And then I need to go here and there. You have to jump all over the place. And you go to the context in the past. And that's how it's always been.

And it's been a big part of our jobs, actually, a heavy lift. And it takes people to do that. And that's starting to change. And so model context protocol.

Hopefully, this is a visual that Tori created, my product marketing manager with Renee. And I think it's really helpful to understand just like, MCP, we hear that a lot. What the hell does that mean? And what we're looking at is the model.

So you have your LLM model. And then you have your context, which is sitting in all these systems, those things that we used to have to go and find ourselves. And with AI, like a year and a half ago, or even a year ago, it's like, go to that system, copy and paste, put it into your LLM, have the LLM do its thing, copy and paste, and put it back into another system that you want to use something with. We've been like that middle piece connecting things.

MCP is a protocol that brings the context from your systems to the model so that you can work with them. And it also allows you to push back to systems that have the right APIs open and everything around that, too. So it can write, you don't have to copy and paste anymore, hopefully ever. But I know in some systems you have to still.

So it allows you to take all the tools in your pack that you carry as a business professional and start to connect them into this intelligence space that you get to work with. And it becomes your partner. And you can bring it in and pull all of that context to you. So rather than going out and finding it and gathering it yourself, you ask the model to gather that for you, pull it all together, and then it allows you to work with it.

So that's kind of the gist of an MCP. So a little bit more. So we're going educational first. Your tools have tools within them.

These are like either APIs or they're different tools that are added to an MCP that allow it to do certain things. So like Gmail, you can search threads. You can get an individual thread and pull the full conversation. You can create a draft with it.

With MCPs, you can't send an email. With Gmail, you have to create a draft, and then you go to your drafts, and you can edit and send it from there, which I think is actually a good choice. It's definitely a human in the loop choice in that way. So that's an example with Gmail.

I use linear with my engineering team. And so when I'm working on product features and ideation and things like that, once I've whittled it down and I'm like, yes, this is what I want, I push it to linear. I don't have to log into linear. I don't have to go type anything in there.

I've done all of my work with the LLM, and then it creates my ticket for my engineers for them to build a feature for you guys. And so all of the tools have these different sub-tools that give you capabilities of how you can use those tools with MCP and how the model's allowed to interact with them. With our MCPs, so both the Gainsight CS MCP and the Staircase MCP, and there are more MCPs on the way from Gainsight, I just today started playing with our PX1 for staircase usage. I'm pulling that in and experimenting to push the edges of the MCP and figure out what it can do.

I know Skilljar is also working hard on their MCP as well. So all of our tools will have MCPs. You'll be able to access them. Today we're focused on the CS tools.

So with Gainsight, we have a very different structure than we have with the Staircase MCP. Gainsight has a lot of APIs and external APIs that you can expose and use with different tools. And so the MCP for Gainsight is built on those APIs. It allows you to use those APIs to read from things, to look at success plans.

And then it also has a few write back skills. So it can write to timeline. It can write to success plans. It can create a CTA and update a CTA as well, which is awesome news, I think, as being a CSM and juggling a swath of CTAs that I don't want to close myself.

And so it definitely can accelerate people. Staircase, on the other hand, Staircase, we actually don't have external APIs exposed. And what we're doing is building an agent over context. And so we have an agent that we're giving tools and capabilities to look at our reports and tables, to search for different things, to pull in, we'll be adding our events into it so you can search our events.

All of the different things within Staircase, we're training the agent in Staircase to know how to use these things so that when you ask Staircase a query, it will decide these are the tools I need to use for it, gather the right things and send it to you. So different models, but similar outcomes. But they do kind of function and feel a little bit different also in the outputs. These are some examples of the tools that are available in the Gainsight and Staircase MCP.

These are all, if you connect the MCP, you can ask it like what the list of tools is and it'll tell you. So it's not like this is like secret information or something like that. But you see there's a lot of tools available there within Gainsight where with Staircase, we have a query, we did just add the Staircase generate report, which allows you to definitely increase your ability to do cross account analysis by filtering for a CSM and pulling things and we'll model that and how that looks. But these are, we'll continue to add tools into our MCP, open up more capabilities and give you controls on what's available and possible for different role types and teams, especially within Gainsight for write back permissions and things like that.

So my experience as I've been building with these MCPs is one of the things I like to do first is push the edges of what's possible with this MCP. And it's not me doing most of the heavy lifting actually. By show of hands, how many people use Claude in the room? My family, I love you.

So I've been a Claude fanboy for a while. It was two years ago, July, that Claude and I started our relationship and it's gone pretty damn well, I'm not gonna lie. At that time, I just really liked the way Claude felt and thought and felt like it got me a little bit better. Wasn't like too different from chat GPT at that time for me, but it just like, I liked what it did a little bit more.

And it felt like I could give it some personality and I could do things that really worked for me and resonated. It also had things like, it was the first to have canvases and artifacts and projects and other things like that. Anthropics been very innovative in their tool sets, I feel like. Then the last, so it was like last October, I was getting really, really jealous of Claude code people that were using it because Claude code was iterating so fast and releasing all these things.

And I was like, I want that in the desktop app. I want that, I was like feeling really like FOMO a little. So I had Claude set up Claude code for me on my desktop and I started using it in the terminal and I built out this whole system. And it was a heavy lift, I'm not gonna lie, to build out like Claude code for knowledge work rather than for coding.

I don't code, it's not something I have, I've never written a line of code. But I do have a full like architecture built out that I've thought through and built with Claude for my knowledge work. So I've been using Claude code now for seven months and now the desktop app, I'm like, ooh, co-work's really cool. And actually Claude code in the desktop app is really good.

I'm starting to migrate back to the desktop app now that the features are more live there. So that's a little bit about my Claude history. But Claude is my closest teammate, I'm not gonna lie. I spend more time with Claude than anyone else that I work with, hands down.

And it's smarter than me now. That's another thing I've seen about these models is like two years ago it felt like, ooh, this is a pretty good high school graduate. Like they're smart in that way. And then a year ago it's like, ooh, this is like college graduate.

This is cool, it's like I have a good intern working with me that I can tell to do things. But I have to correct it more, I have to keep it on its rails and do these things. And now it's like, ooh, I just get to brain dump and tell like, ooh, this is all the things that I'm thinking of. And then Claude creates the plan, Claude sees up all these tasks and it's like, yeah, you want me to go to work?

And I was like, change this, change this, change this. Yeah, go to work. And then it'll go to work for 15 to 20 minutes and do a bunch of different things for me and then give me outputs that then I start to iterate and work with at that time. So it starts to be a little bit like juggling agents and like playing with things and jumping tabs.

And then it's like, oh, now I go do this work and I come back to it, especially for longer form tasks that the LLMs can run in that way. So with MCPs, you can, like any MCP you have, you can tell Claude, and I'm just gonna reference Claude, but if you use a different LLM, I apologize, I know you've probably heard Claude a lot too. It is the tool I use most often and I do find it to be the most agentic right now because of its harness, how it uses the tools that it has access to. Is this too much or no?

No, okay, all right, good. So when you get a new MCP, like tell Claude, hey, I wanna test this MCP, I want you to dig into like the capabilities, what are the things, these are the things like I think I wanna do with it, what else could I do with it? Go in, dig in, push some limits, grab some stuff, see what we get, how is that? And then let's like build a plan for how we're gonna work with this going forward.

So you can work with Claude to push the edges of the MCP is what I like to call it, to understand it and to build skills around it. So all of this, actually a lot of what I just talked about is context engineering. Like all of, and we've moved from a place of prompt engineering being really, really important, like how you prompt the LLM was very important a year and a half ago. It determined a lot about what happened.

Now, like I said, I can brain dump and I just like, I'm just filling it with as much context as I have in my head and filling out those context edges. So it is like, what does it need to know to be able to do the thing that I want it to do? That's context engineering. And all like the file system stuff, projects in Claude or other systems, you can add like files to it and give it the context it needs to be better at its job, to be better at your job actually.

And so context engineering is now the work. Prompt engineering is not that important because the LLMs themselves are really great at like understanding things and crafting their own prompts. So with MCPs, what I've learned is I've been really exploring with the GainSight and Staircase MCPs as I see like three layers. We have those tools that I talked about, like what can the MCP even do?

What does it have raw access to like hardware essentially? It's like a USB-C plug that you're plugging into things. Then what I've learned is I also need to add a foundation skill onto this. So what do I mean by that?

And it's teaching the model how to use the tool that's there, right? So I kind of described that like push the edges, figure out like how we use this. And then once you figured it out, now let's make a skill. You tell the LLM make a skill so that we know how to use this in the future every time we use this MCP.

So that's like that foundation skill is how do I want it to use that? And then there's workflows on top of that. Once it knows how to use the tool intelligently, then it's like, what do I wanna do with this? I want meeting preps.

I want meeting processing after the fact. I wanna be able to update and manage success plans and CTAs. And so then it's like, what do I wanna do? And how do I make that a repeatable task that the LLM can take going forward?

With skills, first the metadata is the trigger. So you can hard code like if I do a slash command, it'll do something. But you can also just have metadata in the skill that says if I say like, hey, process my meeting, then it'll automatically, oh, I need to use this skill and pick that up and do it. And it's getting better and better at doing that.

The other unique thing about a skill is progressive disclosure. So this is what really makes skills unique compared to just like a really long prompt. Before, that's what it was, is you'd have like a page long prompt with like all these details and all this stuff in it. But that takes up a lot of context, not that much, but it depends on the prompt length and what you're adding to it.

That takes a lot of context and it's expensive, honestly. It's expensive and it's energy inefficient as well. With skills, it allows for progressive disclosure. So you'll see like at the top level of this, it has a readme file.

And that readme file has instructions for the LLM that tells it, hey, these are all the things that are in here, this is when you should use what. And so it reads that first and then it's like, oh, for this, I actually only need to use the query patterns and I only need to use the meeting processor skill. I don't need to use the account handoff, I don't need the daily cockpit. So it does not pull that context in, doesn't waste those tokens, and it's more token efficient, and it ends up being better at what it's doing because of it too.

So that's kind of the unique aspect of skills. I will say Anthropic has made skills open source. They invented MCP and they made that open source and now that's been adopted by all the models. They've made skills open source.

I have not seen the other models adopt it yet. I hope they will. It has been open source since like December or November, I think. So I think I predict it'll be a foundation for all LLMs.

So I'm hedging my bets that building good skills will pay off for other LLMs in the future too. This is just best practice. Anytime you're doing something, always ask like, can I do this with my LLM first? And then do it with your LLM.

And then the thing to ask after that is, can I turn this into a repeatable skill? Is this something I need to do often, regularly? And if I'm doing it regularly, then after I've kind of tuned it, the first time you do something, it's a heavier lift. You're feeding it context.

You're massaging the output and how it should do things and steps and things like that. And then at the end of it, you can just say, I want to turn this into a skill. Go, turn it into a skill for me. You could give it a little more guidance if you want.

But then you create a skill. The next time you go to do that thing, it knows how to do it already. You don't have to massage it nearly as much. It takes a lot less time.

You can refine that skill if the next time you run it and you're like, ooh, I don't like this. You can tell Cloud, hey, update that skill. I don't like that. And it'll update it and fix that for you.

So it's a model kind of building its own structure with you and guidance. We're almost into the workshop. So the last thing on this is this is something I see happen in organizations. And I promote it.

I think it's something that should happen in organizations. It is like every individual is working with their LLM and hopefully creating their own skills and processes and things that they're using it for. Those things can then be compounded into value for teams. If somebody has a skill that they're like, man, this is saving me so much time.

Man, this is awesome. As a CSM or as an admin or as somebody else. You can then just take that skill and say, hey, I made the skill. And it's a zip file.

And you share it with people. They can install it. They have superpower now that you created for them. So the ability for users to build is very high right now.

And it's very possible. It's a little bit vibe-cody, but not vibe-coding, really. But what I usually see in organizations is there are a few people that just love playing with LLMs and love playing with AI. And they kind of sprint out ahead.

And they build stuff. And they're like, this is awesome. This is so much fun. And other people are like, this is not that much fun.

But the people that are sprinting out ahead can be like, hey, here's something I built that I think will work for everybody. And so use those outliers as the guides forward in an organization. And maybe even give them a role around that would be my recommendation, honestly. I recommend keeping them in a role by giving them like half as much work to do so that then they can spend the other half of the time building stuff for other people.

Because if they're still doing the job, then they build stuff for them to do the job better. So then that scales to orgs. And also, it's scaling the products. Like I view it as my responsibility to build generalized skills with Staircase that give value to you as a customer with our MCP.

Like that's on me. I know Staircase better than anyone, actually. I'm pretty confident about that now. And so I can think about creative new ways to use it, build a skill.

I don't even need engineers to do anything, maybe. Maybe I do. Maybe I'm like, I need access to this. They can give me access, and now I can build a new skill.

So I think products will start to ship a lot more plug-ins skills and make it out of the box valuable for people. So you don't have to build it anymore is the end goal. But right now, we're all kind of learning together. So I'm going to transfer and switch to Cloud.

Hopefully my laptop will start. [SIDE CONVERSATION] Arc needs an update. That's good to know. And this is my terminal app where I work with Cloud.

I'm not going to work there because it's too scary for most people. OK, so I'm going to do a few live, but it does take a couple of minutes to run this kind of thing. So I'm in co-work inside of Cloud right here. And actually, I want to show you a couple of things first.

So the plug-ins that I just showed you inside of GitHub. If you have Cloud, you can go into Customize over here. You can go to Browse Plug-ins. There's a few steps here.

You go to Personal. Your organization can add them if you're on Enterprise or Teams. Anthropic has their own. And then there's Personal too.

These are some that I had built before. But then you can click the Plus button and add a marketplace. This is how you would find the ges-bbloom backslash Gainsight-mcp-plug-ins. And that will just pull that in.

You sync. And then you have-- I only added the Chef Gordon here in the Fun Life Skills. Maybe I'll add more over time. But then you have the Gainsight CS one.

This skill set is specifically focused on using Staircase and CS together. I will be working on just Staircase standalone ones. I've learned a lot about Staircase. And actually, there's a few in here if you see.

There's Staircase MCP Expert. So if I go in, this shows you the skill structure. These are all the skills that I've built for the plug-in right now with some workflow ones and a couple of expert ones. And so this expert is like, hey, when do you invoke Staircase?

What are the top patterns that you can do? How do you do cross-account queries? How do you do individual account drill-downs? So it's teaching Claude how to use Staircase.

And so that skill is available. And then it even has these references that has more context around it. Ooh, what are analyst data models? It has the data models that are available for it.

So again, this is all the instructions that you don't ever have to read. I have not read these yet, to be honest. Claude and I worked on it, and Claude made it. But it makes Claude understand how to use Staircase well.

Now that I've shown that skill structure, we'll go back in. And let me kick off a few, and then I'm going to show you ones that I preloaded. So I'm going to work in a project. I created a demo project, so it just has a little bit of demo context in it.

And now I'm going to say I want to run the CSM book pulse for Hannah. So this one is meant to be, like, if I'm a CSM, I want to filter for all my accounts. I want to pull just my accounts into context. And then I want, like, what should I do, essentially, with my accounts.

And now it's going to start. And it's going to do a few things. It's first, like, thinking about, all right, what's it asking? It's running the skill.

So you can always click here. And it says, oop, it's running the CSM book pulse. And so now it's loading all that context about the CSM book pulse that it needs, the references and things into here. It's loading all the tools.

It says it's using tool search. I'm going to let it keep-- so it kicked off a Staircase query. We can see what that is by clicking it. List all the accounts where the CSM is handled.

For each account, include ARR, renewal date, health score, last engagement date, last reach out, sentiment score, engagement score, risk level, expansion readiness level, and indicate whether the account is currently flagged with any of those insights. I'm going to go bigger so you guys can see. With any of these insights, account dark, no QBR, no reach out. And actually, I've learned the insights aren't quite available with our MCP yet.

So it'll probably get mad at me. Not mad, but it's not Gordon. But it'll tell me that it couldn't access those things in the output that it does. Now, it's also pulling context from Gainsight in this example to gather-- and are there any open CTAs?

Is there an open success plan? Have there been any recent timeline updates or things like that? It's pulling for that context now, too. And then it's going to start thinking and working.

So it's gathering context. It's pulling in those things that it knows it needs to be able to run the CSM book polls. And then it's going to keep thinking about it. While it thinks, maybe we'll come back and look through the steps.

Which, by the way, if anyone wants to, you could try this if you have the MCPs. And you can install them in the plugin. It's supposed to be a workshop, but this is kind of a little bit harder to workshop. I ran a cloud workshop in February for our teams, but I knew they all had access to cloud.

And I knew they all had access to our MCP. And they could get in and play with it. But hopefully, this is inspiring you, not only for our MCPs, but others that you have, too, and how to use it. So it is a little bit more model-y than that.

I am open now that we're in this portion, too. Just like, if a question sparks up, throw them at me at this point. I've been talking too long, anyway. So feel free to throw questions out.

[INAUDIBLE] I have added a skill inside of the plugin that teaches it how I like it to use the visualized tool that Claude has. So Claude has capabilities to visualize and add little widgets and buttons and things like that in the display and build it on the fly. It will do that. But as I was having it do that, there's things I wanted.

I wanted it to behave in certain ways. And I wanted the output to look in certain ways. So in this skill-- and you guys all have access to it, so you could actually just even steal just that skill and load it into your Claude. There is a skill specifically about MCP app design or something like that is what it's named inside the skill set.

Yeah. Any other questions before I jump into a different example while we watch Claude work? So fetch CTA list, fetch success plan list, it resolved the user and gain site. Now it's thinking.

So I think it has all its context. And now it's starting to think. While it thinks, we can go over here and see-- so this is a meeting processor. So you can pull as long as you have a transcript, which if you're not recording your meetings and getting transcripts and using them with AI, highly recommend it.

I think it is hands down the most valuable use case for anyone in any job where they're joining Zoom meetings or Teams meetings or anything like that. Get your transcripts and use them. So it asked me first, what type of call was this? And it was just a cadence call.

I told Claude before. And I gave it a little bit of context. And then it pulled the transcript from Notion. I have Notion meeting connector.

So pulled the transcript. This is a mocked up transcript because I don't want to show real customer data. So it is looking at our demo works and stuff like that. You can see here it loaded tools first.

So it's thinking. It loaded a tool. I think probably Notion there. Now it's doing lookups.

So it did Notion. It looked up Staircase. It did an account lookup to get context about that account. And then it also resolved the customer.

Within Staircase, I think it was doing that. Oh, no. It was Gainsight that it did that. You can see all the details of its thinking if you want to click in.

But Claude does a good job of hiding stuff that you don't need to see unless you want to drill in and look at it. And so I didn't do anything other than say process that meeting and say, yep, that's the right transcript. And then down here, there's a meeting recap. So I actually got this format idea from Kalpana Krishna Kumar.

She's one of our CSMs. And I met with her a couple weeks ago. And I was like, hey, how are you using Claude? And she showed me one of her outputs.

And I was like, this is so cool. What a cool tool. And I was a little jealous. So I built it for myself, too.

So it starts with a summary. Here's their health score sentiment, the days to renewal, how many open CTAs or success plans they have. Then it has here's the action items from the call. This is a high priority action item here.

And then there's an expansion and a renewal happening. And we need to have an EBR plan based on the conversation. It even pulled out a quote from them. I think it's an advocacy quote that it pulled out.

And you see down here, it says, there's two recommended rights to gain sight and one email draft queued. I could just click this and tell it to go. But I want to look at them. So here's the follow up email draft.

It drafted this. And it's, hey, Travis and Nicole, quick recap from the sync. Here's the recording. Here, optimization, bulk import.

It's whatever. It wasn't a real meeting. So I can improve and create this draft right here. And then I think Claude kicks in.

And it will use the MCP, drop it into my Gmail drafts. And it'll be ready for me to send. Gain sight. It shows, hey, you should post a timeline activity.

I've drafted it for you. Do you like it? Are you ready to go? I could tell Claude to change it if I want, or I could approve and post it.

You need to open a risk CTA around their renewal right now. So here's the tasks for that that I've set up. Do you want to edit it? Do you want to approve and drop it in the cockpit and send that off?

So all from that meeting, it's teeing me up. That's what I'm always going for, is teeing people up to knock out their work. [INAUDIBLE] Yeah. And it's there.

So then why are we creating another timeline entry to summarize the meeting that it just summarized? Good feedback. And then it's going to be there for the next time it runs, and it's going to see two. I think what I would optimize from this is-- because again, this is preview right now on my skills and stuff that I'm playing around with.

And I haven't gone through individually each one to refine them yet. It was more workshop prep. And so what I would do on this one is instead of a generic summary about it, because that's already flowing in, is there any key thing or update that I want to highlight as a CSM update instead of a meeting note from it in that way? Or maybe the optimal would be-- I don't think it's exposed right now, but the ability to add a comment to the timeline of the meeting, and then say, yeah, let's do that.

Does that make sense? Yeah. But yeah, that's a great example. And maybe it's possible, but I don't have to post this too, because I know the meetings are flowing there.

But it would be ideal to optimize this. Look for anything specific that should be called out as a separate timeline. The summary is already dropping there. I could train Claude to know that.

And then it could function differently. Yeah, it's a good call out. Any other questions as I'm going through a meeting processing tool? Yeah.

Thank you. I think the feedback there is instead of timeline activity, CTA tasks. It's like the functional component of what are my actual things after this. Give me a time frame to do it in.

That's kind of like the getting into the system's design kind of mindset, where working with admins or working with your CSM, whoever, to kind of come up with this next version of this. Here's kind of where our preview version is, but the functional part of that. This is really cool. I love the idea of give me my next set of tasks.

I'll log it as CTA, and then Claude can remind me of it. Next time I have my daily digest, here's the thing that we did yesterday, something like that. Yeah. Yeah, 100%.

Kalpana, in her flow that she showed me, she had real success plans open and real CTAs open, and she had a meeting with a customer and pulled it in. And so hers was not draft a timeline entry for the summary. It was like check the CTAs. And it's like, hey, there was a verified outcome you talked about on this call.

Log it to the success plan objective as a timeline entry tied to that objective about that input in that way. This is stuff that like, guys, I've been a CSM, and I only had 10 accounts. And logging all the stuff for just 10 accounts was a pain in the ass, and I didn't do it most of the time, is the truth. I'll admit that.

And so when you don't do something like that with a system of record that you want that fidelity, it's like, ah, you lose value. You lose the context value. So the ability to speed that process up and automatically update your CTAs-- hey, you talked about this with the customer. They have a risk, and you had these tasks on there.

I saw that you did two of those tasks. Want me to mark them as done? Yeah, please do it, and update it like that. I think we will build towards this with our integration between the products, too.

We're in our meeting processing. Now that we're more agentic, I want to go back to our meeting processor and think about it in this agentic way. Hey, look for success plans and CTAs first with the CSMCP and see if there's any updates that should be recommended. We can build that out of the box, but it's possible now, also.

You can build it right now, too. And so there's certain flows that I think we'll build. There's certain flows that I think you should build with LLMs and MCPs. Here's some wins that they had.

There's an advocacy quote. There isn't a thing called advocacy library, so I need to teach Cloud that. And then there's some just like briefings on things that are happening with the account, also, to be up to date on it. Go ahead.

I'm curious on when you're building the skills, do you advocate for the skill builder that Cloud has-- The skill folder? The skill builder. So the Cloud has the skill builder skill. Do you use that when you are building and refining?

Or do you-- Yeah. Do you build them yourself or-- I use both the Cloud skill builder, and then I think there's one in the superpowers repo that exists that has some good things for planning, also, and other stuff like that. So I do have the superpowers repo. I think those are the only two that I've installed.

Other than-- OK, so you use both? Yeah. Yeah, I do use both. I have them both available.

I let Cloud use what it wants to. I didn't know if you were building them yourself. Yeah. No, I haven't personally built a skill or custom instructions, and probably a year is the truth.

And so I'm not like-- generally, I'm scanning the plans and reviewing the plan Cloud makes and giving it as much feedback at that early context stage so that I get all the potential errors out in the beginning. And then I let Cloud go, and I let it roll. I actually use bypass permissions mode most of the time, because I'm not building code. And so it's not like production code that's going to break something.

It's just my markdown files. Like, it's a markdown system that I have. Cool. That's what I've done is build NfPad, build the skill in addition to building markdown so it does everything in one prompt.

Yeah. I'm using Obsidian as a tool. So here's Obsidian. And this is such a bloated-- there's way too many files in here.

Every month, I need to be like, all right, let's go clean up all my stuff. Because I'm working on making it more self-automated. But this is like my Cloud brain, essentially. And it's all just in your file system.

So with Cowork and Cloud Code, you can just build knowledge in your file system on your computer, and you point it to that folder. And then it can use all of that stuff as its brain system, essentially. Started with code that way, and Cowork makes that more accessible for everyone to use in that way as projects. Oops.

Dang it. I didn't give it permissions. So it was asking-- oh, it was asking to draft there. It's asking me to do something here.

Oh, it's ready. So I ran the CSM Book Pulse for Hannah. So this is looking at all of her book size. She has 31 accounts.

They're all SMB tier. This is her total ARR. There's four high-risk alert signals. There's one overdue CTA.

There's five EBR she has upcoming and seven are expansion ready right now. Nine accounts are past their renewal, because this is demo data. And so the demo data is not always the cleanest. It's pretty good demo data.

I'm not going to lie. But it's still not perfect. It's not real. And then this one, in particular, they renew today.

And so you can just jump straight to the account in the C360, because it can link the account page. Here are your priorities as a CSM. This is something so cool that I'm like, I love the display on this. So here I have general information.

But watch when I click on Rosales Technologies. Pops up. This is their state. These are the next moves that you should do with Rosales.

I click here. So each one of these has-- it's ran thinking for all of them and prepped this for me so that I could take some actions and I could do something for each of these. Draft a check-in, reach out, et cetera. These are some of the types of things that an agent studio will be able to do more natively inside of Gainsight with more walled restrictions and permissions for users that use admins or organizations will have more controls over to set up for your teams in that way.

So those are priorities. You have active work open here. And here's a briefing on all of these. And you can jump to any of the pages.

And here's a high-level briefing. There's another cool one. Am I almost at time? Or no?

Yeah, I am almost at time. So I'm going to jump back in. But here's one like an exec view where like, hey, what are my upcoming renewals? And this is priority ranking based on risk and health.

And then again, you can click into each of them and you get like a mini summary with their states and status across all of these. And I can click through and see all these things. So trying to think of different use cases for different users of our tools-- admins, leaders, managers, CSMs, account managers-- and be able to build those. Here's cross-account themes that we're seeing.

And here's resource allocation over these risk accounts, like who's managing them and which ones need help versus they're doing fine. Let's go back to the slides. So that was kind of fun. I had a good time if you didn't.

As long as one of us is happy after this session, then even if that person is me. So now it is your turn to build too. Thinking about those different use cases, I would say especially as like admins and leaders, but for individual contributors, think about your use cases and how you can use it in that way. As a manager, think about your use cases and your team's use cases too, your individuals and how you can build for them.

Think about the tools you have access to and what you can do with those different tools, which ones are valuable to pull in. If you're going into a system all the time and you don't have an MCP approved, that's available. Talk to security, move through the chain to get it approved. There's all kinds of blockers to moving forward in organizations to these things, which there should be.

But going through the proper reviews and getting that done, then you can start to play and build. So start to play and have a good time. Here's the core takeaways. Oh, yeah, I guess my first core takeaway is play.

Get your MCPs connected. Build some skills. Play around with it. Get your organization to buy Claude if you haven't yet.

Anthropics should pay me like a commission. Play with the connectors you have available. And create one like real task. And at first, there's friction the first time you build it, but then it smooths out and gets better and better.

Ask and codify. So can my LLM do this? Do it for yourself and then share it with teams. Make things repeatable so that it lightens your job over time.

And you can either do more or you can take time off, hopefully, which I don't get to do. And then coordinate also. So I think coordinate is something, if you are an organizational leader in the room right now or an admin, it is on you to organize around this. Your teammates will sprint ahead and they'll be doing stuff under the radar most likely.

Or maybe they'll try to get attention and be like, hey, we should do something about this. But if you don't listen and if you don't give them the space to do those things and to learn together and to create an AI club that everyone can join and share what they're doing, create a role that focuses on this and helps to get things organized, it's up to us to help individuals learn how to use AI and to learn how to use AI ourselves too. So definitely organize and get coordinated. And we've done some questions along the way.

We're out of time right now. But I'm around-- actually, I have a meeting right after this. So I'm not around. But catch me really quick and I'll see.

Actually, no, the customer would be really mad. I have to run right now. Thank you.