it's 2 after and I think we have a quorum and hopefully people will keep joining but I think it's time for us to get started Um I am Retesha I live in pure sometimes cold Michigan with two beautiful ladies married to one for 17 years and then a 10-year-old daughter who acts 16 often By trade I'm a geni architect at Service Now Prior to that I was a gen specialist at Amazon and I've worked with customers like you all on AI implementations and hope to share some real world experiences here today Ashish thank you Resh for the wonderful introduction My name is Ashish Si Uh I live very close to Riches not a long distance Midwest Windy City Chicago And I also have a 10-year-old daughter and uh who definitely acts like 16 for sure maybe 18 as well able to manage my son who is 6 years old and I am from the now and now team within service now for those of you who do not know about now and now we are also known as customer zero which means we use all our products internally give firsthand feedback to our product teams we co-design co-uild stuff with them so when it reaches out to you our customers partners you can get maximum value out of it looking forward to our discussion today and hey everyone I'm Tara Fischer um I'm here in sunny California And um I also have a I actually have a teenager She's 14 uh going on 22 Um so that's always fun and exciting And um I work from in outbound product management And what that means is I'm here to build relationships with you and really get in and figure out what you think of our products Um if we're headed in the right direction and if you've got any feedback I am open and always welcome to it So please drop my name down and send anything my way um if you have something to share Excited to talk to you today Awesome Uh we're going to share real experiences but just in case safe harbor if we have any forward-looking roadmap items that we talk about today Um this webinar is part of a series that is led by Service Now experts that help you deploy products in your instances Um so real life experiences are shared Hope you get some value out of this today and love this Uh if you want to join future ones you can scan the QR code for the schedule or there will be link in the chat as well And I promise the last housekeeping slide uh hopefully the most important one Uh we love your questions This is for you Please throw as many questions as you can Use the Q&A in the Zoom panel And you know it's good for me to say that I don't have to answer anything Terra is going to answer all of them for you today Um some of the answers might be let's ask Retesh so be ready for that We'll do our best to get you the answers today We'll also have time for Q&A Um so you know we'll have that If you're like me and you know nerdy take screenshots and notes Uh take it easy All of this will be recorded put on the community so you will have all of it And then the last thing a favor Please please please at the end you will get a survey We love your feedback If I sucked it's an opportunity for you to tell me that so that I get better Um but I hope to hear that I did well today All righty let's set the stage a little bit So AI agents a loaded term confusing all over everybody's talking I'll be honest I have read multiple white papers done a ton of hands-on and I still feel like I don't get it So if you are in that state of mind it's perfectly normal This tech is evolving so fast we can't humanly keep pace with it And you are in the right webinar by the way So Ashish is going to get us started He'll demystify AI agents what they really mean in the service now world Not only that uh we drink our own champagne and proudly so So Ashish will talk about how Service Now uses AI agents internally driving $325 million of annual value He'll also show you a demo So he's you know he's all set up to do all the fun cool things to get you excited today And then I'm going to crush the party uh mellow down that excitement take you to the back of the kitchen uh show you some ingredients and recipe Hope you want to see that Uh we'll talk about architecture building blocks uh some leading practices and then the most common question I get from all customers how can you help us or what do we need to do for a successful implementation so we'll talk a little bit about that and I promise we'll keep time for questions answers today Um just one last thing so we get a pulse of where you are at Uh you should see a poll Uh we would like to know like where you are at in your um AI agent understanding or journey just so that we can adjust our content today and then you know deliver value for you There we go We'll give it a minute or two So uh while we're waiting for this poll and and see the answers coming in it looks like folks are somewhat familiar with this Um Ash I'd just like to put a call out to you Um our very familiar is pretty low So if we can make sure that we break this down a little bit so it's not super complicated to understand I think um that's kind of the audience that we have right here is it's the shiny new toy and we want to make sure that it's consumable for everybody So we can get more forks uh more forks more folks using this sooner rather than later So yeah we have a good sample set I'm going to end the poll um and you know you can see the results but 50% of you are in between which is like the perfect audience perfect content for today Enough of all the housekeeping and gibberish I'm going to give it over to Ashish Thank you for setting the context I think we all mentioned that we'll start with understanding a agents Glad to see that you know people have read something about it know the theoretical concepts on it but still trying to learn more and trust us that we all are learning almost on a daily basis as it mentioned hands-on read multiple articles white papers still see something is missing am I not understanding it fully so we'll try to break it down but I'll try to you know take you back to you know five seven eight years back we didn't start with AI yesterday oh ji let's jump onto AI service now as a platform as a company has been a believer in AI for years If you know about or if you remember long time back we had predictive intelligence already existing on the platform The four aspects of it classification clustering regression similarity all of that was playing when it comes to predictive AI So these are artificial intelligence systems that analyze historical data identify patterns forecast future outcomes It obviously leverages machine learning statistical modeling data processing data collection The more data you provide the better prediction the better outcomes are Then comes your perception based AI These are AI systems designed to interpret and understand sensory data such as your images audios and other inputs This is where your natural language processing natural language understanding made it easier for computer vision And that's when your augmented reality virtual reality all of them came to life So this has been also there for a while Our platform has been a big proponent of natural language understanding through virtual agent If you if you realize that the chat bots and everything and since last couple of years we have been riding the generative AI way We all have heard known about genai genai Now assist is our large language model which is backing up the genai way These are obviously we know that AI systems that create new content such as text images even code based on the patterns lo from vast set of data Unlike traditional AI uh what makes GNI unique is that it can produce original content So it's trained on vast amount of text images data based on that it can generate more content And since last late late last year we have been riding the Agentic AI wave So Agentic AI these are AI systems that can autonomously plan execute tasks and adapt to new situations with little to no human intervention Unlike traditional AI which primarily responds to inputs passively agentic AI is much more proactive Goal directed behavior based on reasoning planning and adapting as well because it's context aware And as you can see you know by a prediction by Gartner it's a since last year 2024 there were 0% routine decisions which were made autonomously by Agenti that will rise up to 15% in a matter of four years in 2028 it will be 15% So huge game changer for us and can anybody predict what next wave of AI is going to be right yeah If if you believe me the way I've been studying and being close to this I think the next wave is always going to be about completely autonomous AI Right now it's still human in the loop still supervised because we still are a bit jittery Hey it's going to do the things in the right manner or not So let's seek what the future holds for us Let's ground us with some concepts Ashish like basics Absolutely I want to do a deeper dive into this and all about this So 2025 is being termed as a year of AI agents This is the new frontier gamecher This is obviously a latest buzzword on the lips of executives uh of large tech firms and obviously their clients And I was reading a few facts uh in an article as well Amazon increases its productivity and uses 30,000 uh AI agents to upgrade its 30,000 production applications saving the e-commerce giant around 260 million globally Another prediction by fund manager Arc Investments estimates that AI agent will enable 9 trillion which is 25% of e-commerce sales by 2030 in just a matter of five years So what's what makes them so powerful and why is everybody talking about them few characteristics or traits of agentic AI listed here Autonomy It can make decisions and take actions independently Goal oriented instead of specific task focusing on a specific objective It is understanding and focuses on the larger goal It has planning and learning capability It can plan those steps and it can learn It has a short-term memory long-term memory It can learn as well It can adapt It is context aware It knows about your environment Retains the past experience as well As such it can adapt course correct as well And it has multiple tools available to its disposal So these AI agents they use workflows flow actions subflows and any capability on the platform they are able to use that The that makes these AI agents much more powerful And these agents are able to collaborate amongst themselves and get the job accomplished So all of these traits and characteristics make these agentic AI much more powerful today and that's why everybody is talking about them So Ashish let me ask you a question here because I've actually had this uh question come up from other customers if we already jumped on and started leveraging some of the things that service now offers for Genai which is our our now assist brand um is that now with a AI agents coming into play is that throwaway work or can we com combine them like can you explain that absolutely I think this slide should perfectly answer and that was the intent of putting this slide together so all the ji work now says skills you're talking about NAS now says skill kit you have built custom skills There is case summarization incident summarization all the geni capabilities which you have available all of them will make these AI agents powerful So all the work which you have done all the workflows that you have built on the platform will absolutely be used by these AI agents to make your business even run faster and much more efficient So let me take an approach on how to explain these workflows So this slide always reminds me of you know uh my 10-year-old daughter asking me "Hey dad what does service now do?" And my wife always you know nudges me hey you are in the platform for like 11 years master architect Why can't you explain it to a 10-year-old daughter and I always explain these technical concepts to her And she says I still don't get it And then I steal one of the lines which our CEO Bill McDermott says we make the work flow That's why we are a workflow company So workflow is exactly that So they are predictable rule-based processes If this happens then do this If else do this So workflows will still be in existence Anything which is static static in nature deterministic and outcome will obviously be done by workflows When it comes to speed much more reliability workflows are your go-to go-to aspects and then obviously they do repetitive task and that's why they are much more cost effective When it comes to AI agents they are real time reasoning There's a multi-step reasoning and they make decisions on the fly as well based on the input and the context They are much more adaptive and non-deterministic which means they have the boolean logic as well as a fuzzy logic which a bit with a bit of autonomy built in in it I'll give you an example Let's say if I have to make a reservation at a restaurant I will know use my GI capability I'll say I want to use this uh um capability to make reservation at 8:00 p.m uh for a dinner for 2 And then it's going to go look into the menu of the restaurant made the reservation and provide the details to me And uh if I ask an AI agent to do that hey I need to have some dinner plans to be made at 8:00 p.m today it's going to know my location So the way it's going to act is that it's going to check my location It's going to check all the restaurants in the vicinity and then it's going to check the restaurants based on their availability which ones are available open at 8:00 p.m and so on and so forth It's going to also check the ratings the feedback which the restaurants have received It's going to consider my dietary restrictions If my lactose intolerant I'm vegan and then it's going to look at the menus It's going to make the reservation and then it's going to give me the confirmation number So you see how AI agent was different from GI It knew the context It knew where I was It also knew that what are my dietary restrictions and what is the open close time of those restaurants That's how they are more context aware and they can course correct based on the situation as well and that's why they have more and more autonomy as well as reasoning and they are very powerful and comes to complex large task and you will see some of that in examples which we throw you in one of the demos as well Okay So next topic is around service now AI agents What AI agents service now has to offer With the March release we have released multiple AI agents across our workflows We realize the need of our customers which can get them started on their agent journey and that is why you have you will see lot of agents getting released with our Marg Yokohama release For example in the IT space we have generate change request plans categorize incidents generate post incident reviews and then resolve request as well which uses three agents at its disposal Next best action recommener user detail agent and record handler Similarly in the HR space you will see a tuution re reimbursement use case and then uh in the platform specific as well you will see analyze and improve services investigate IT problems So multiple of these use cases which are essentially your business problems and then multiple agents coming to solve a particular use case as well And the best part is that if there is anything specific on your platform just like you have been building applications on the platform you can use our very own AI agent studio and build use cases which is essentially your business goal which rees will do a deeper dive into and then tag multiple AI agents to it They are very simple They needed natural language simple set of instructions and you can build uh you know very quickly these AI agents on our platform Okay Hey Ashish um are you going to be able to tell us how we're using it here at Service Now definitely So what I've done is that we have uh a particular use case regarding software requestment software request fulfillment uh there for you and then that's how we will go ahead and do ret if you can go back to the uh earlier slide of why our platform is a unique proposition when it comes to these AI agents So only service now unites AI agents the data and workflows on a single platform with a single data model providing one unified experience to make sure we can leverage AI across the entire enterprise So service now AI agents don't just assist they do actions and take actions on your behalf team of AI agents work together to orchestrate end-to-end business processes and then data you already have data on the platform already existing there and then the workflows the AI workflows even power this entire thing so you get that entire unified experience on one platform so as I was mentioning earlier all the workflows which you have built all the data which is already there on the platform all your resolved incidents knowledge article everything will be leveraged by these AI agents to make sure AI is being leveraged in every corner of your business Okay So let's look at a particular demo This is our uh software request fulfillment uh use case I'm sure every company receives hundreds of requests in a month for fulfilling softwares Right at many a time user needs these softwares immediately and then sometimes it takes longer to fulfill these software request resulting in DSAT Uh license keys other information is missing Uh there are other resolvers have to wait in case software licenses are not available and we always ask our teams when you are building these use cases We have built this particular use case one of the first one within now and now within or internally within service now uh to faster identify and provision of these softwares better management and compliance to the software as well and then overall meanantime to reduce meanantime to repair gets reduced So when somebody is raising a request hey I didn't get the software or I'm requesting the software overall time to fulfill those request or even those incidents related to those request gets reduced and this is definitely targeted towards the resolver teams who get those request firsthand and you will see in action these five agents next best action agent approved software identifier which checks the approved software in the system license availability check agent and then ad group license provisioning agent and then the notification agent All these agents get triggered either simultaneously or one by one to make sure they are able to fulfill this entire software request And then there is one more agent which I have not mentioned on this one There is a purchase request as well In case licenses are not available it can raise a purchase request as well Let's dive into the demo Okay So what you see here is uh a playground view in which you can test as well as see your actual interaction happening with the AI agent and the fulfiller I'll put in a task This is a software request which I have received Let's go ahead and execute this On the left hand side you will see the now assist panel through which AI agents interact with the resolver group If I am the resolver and on the right hand side you will see the execution of the AI agents as well happening in tandem You can see the approved software and the software request next back next best action agent get triggered On the right hand side you will see that uh this is the request for lucid chart software It realizes it's a standard software so no approvals required there at all It presents the plan that I'll check for the license availability I'll add the user to the ad group notify the user and if need be I'll request a purchase request as well Yes looks good to me It's a human in the loop giving an approval Okay license for lucid charts are available It checks and says licenses are available and then it's going to go ahead and add your gctor to the ad group for lucid chart and it says I have added the user successfully to the AD group So no verification It's doing the job and verifying things for you on the fly notification has also gone out to your director that the lucid chart license has been provisioned for you and that's it That's pretty much it So all these AI agents get together make sure the license check is happening If a standard software is there or not that is getting verified if uh licenses are not available then in that case they going to make sure they can reclaim the licenses as well Different set of agents going to get triggered If it's not able to reclaim the license it can even raise a purchase request So I as a resolver just need to focus on my screen Make sure that I'm able to give those instructions All these agents can independently execute and do these things in an autonomous manner as well But I want to make sure that I as a human being in loop and giving the approval while the AI agent is doing all the activities in the back end So this is what we have seen uh in the last year or so by executing these multiple AI agents What we have seen is that AI has been a gamecher We have around 200 plus use cases in production today Measurement is absolutely critical and we focus on the value and our AI enabled productivity booster as well So we measured by business units or by persona and you can see in the figure that we have been able to drive 325 million plus the figure which reted within service now 70% 76% of our IT support request are self-service without the intervention of a human being and similarly on the CRM side 72% of our customer support request are also resolved or self-service without needing any live help in the finance supply chain supply chain we have 25% productivity increase and in our developer space as well you know we are a platform company we have seen 25% developer productivity gains when these AI agents are available to our developers 3 million hours freed up in our HR space when multiple AI agents help our HR agents as well and then in the security and risk side we have seen 53% productivity with our server patch management process by applying patches updates and fixes to improve security performance and stability on our platform So definitely a gamecher and this is what we have been seeing in service now till date I'll pass it on to Retesh Awesome So hey Riches as you get ready to discuss with us some of the underlying architecture can you I've got a couple questions that have come up um talking about really peeling back the onion and seeing what the underlying framework is like if you can explicitly touch on that And then we also have a question regarding um orchestration in a typical multi cloud setup So um agents created on Snow AWS Salesforce Um are we going to permit um that sort of integration so if you could touch on that stuff while you're getting into all these great architecture pieces that would be fabulous Yeah I'll try my best Some of my favorite questions So let's just get started So I'm going to talk about the building blocks And you know I like to give you a metaphor so that you sink in concepts Um use case Let's start with the use case Ashish touched it but think about you know something like you want to build your new home Okay that's like your business problem you're trying to solve your goal or objective at the highest level Um there's something that triggered it Maybe you're growing family uh changing locations for a job or maybe you just want to lottery or inheritance I wish I win that lottery someday And then you have the orchestrator which is like your general contractor He's like the highle guy who's planning how to build your house He's not getting into the minutia He's just at a high level orchestrating all the things for you And then the general contractor has an army of agents who actually do the work Uh think of your electricians your plumbers your painters And each one of them have a set of tools So drill machines and wrenches and all kinds of fancy tools that make their lives easy So hope that metaphor helped crystallize these concepts Let's talk about what they really mean in service now terms So let's start with orchestrator Um so orchestrator it uses our GPD40 which is on Service Now managed Azure servers Um if you're wondering token size 128K right now you cannot change the LLM Uh it's an out of the box agent that we ship for you uh but in future there will be a roadmap feature hopefully um for you to to for you to have a choice on LLMs Again keep in mind this is a high level guide not in the minutia just doing your initial planning and orchestrating The next component is AI agents Uh we ship a lot of these AI agents uh out of the box to you or you can create your own So somebody was asking can we build our own agents yes you can absolutely build your agents um in the AI agent studio Um these agents um they use React framework or React strategy I should say And what that means is at the end of the day LLMs are predicting the next token or word Uh but React forces them to reason and then act The way it does it is for building your house it will first have to have a thought based on the instructions you've given So thought would be oh I need to first survey the land do some measurements Based on that thought it will take an action using the tools that it has So it has some ways to measure and survey the land So it will take that action It will learn from it It'll have an observation and then it'll have the next thought and then the next action and then a tool to implement that action and then observation So it keeps doing this thought action observation until it ultim ultimately solves your problem that is building a house In real life uh there is no such agent I hope there will be one day Um a little fun fact about um the orchestrator you know at at home my daughter is trying to compete uh with my wife to be the orchestrator I'm not in the league I get to be the AI agent Uh luckily here they don't have emotions And the reason I bring that up is orchestrator um although it comes out of the box you have control over instructions So you can make them behave the way you want with the instructions you give them And of course AI agents you do the same Um Ashish talked about a lot of tools So if you are wondering you know um what happens to the now assist custom skills you've built or you've implemented now assist all of those are available as tools for these agents uh to work your scripts if you love writing scripts like me you have scripts you could do record oper operations and we even have a web search option I had a customer ask me can you just crawl the web and give some information yes we have limited capability for that as well and what about uh what what what kind of prerequisites are there are there certain things that we need to have ready in our environment to get started with this Yeah let's talk There we go Okay Um so yeah you need to be on Yokohama patch one or we've backported agents to XP7 Um I had a customer you know they have an HR department and IT they both service now The HR leader was asking me should I fight this battle with the IT i would say absolutely yes This is one battle you should fight with your IT um licensing please talk to your um account executives or sales team but you definitely need now assist for ITSM HRST or one of those And you have to have an AI agents uh store app role So even if you're an admin you still need to give yourself SNAI admin role The reason is uh not only least privilege access but not every admin on your team would be ready to change the mindset from deterministic to probabilistic And if you want to have like a limited set of team as you're growing your organ and building skills um you need to have this role to create agents And then the last thing I think AI search is needed Um it is enabled by default I think in most instances And then you also need to have now assist panel which is like the now assist interface for your fulfillers Okay let me just uh get into the tool So once you have all of those prerexs um and you log into your instance um you can type AI agent and that will take you to this overview page in the studio So it has like all the use cases and agents that we talked about um and it gives you some helpful steps on what you need to create and things like that So again just remember highle use case the business problem and then it can have one or more agents and then each agent can have one or more tools So let me just show you uh one of these use cases So I just uh randomly picked one use case in my instance Uh this is to generate a plan for an incident or case or whatever it is At the top you have the name description and instructions So this is where you have control on how you program that orchestrator if you will Um a couple best practices Always be in second person Uh keep things simple So here you see it has only one task to analyze that incident number or whatever you tell it limited to two to five There is not a magic number Uh but keep it simple Don't have like hundreds or tens of steps All that minutia can be in the agents And keep in mind the goal of these instructions here is to help the orchestrator to determine the next AI agent or the first AI agent to pick and then we let the AI agent do all the detailed work Okay Next I want to show you um this is what that AI agent So there was a next best action agent here and this is what the AI agent looks like So again you have the name and description You have an important role thing called as role This is your spot to clarify what the agent can do and cannot do Um you have to make sure that these things are non-over overlapping otherwise the LLM can get confused And then here you know these days these instructions are you are telling the agent what to do Um so you're actually giving it step by step I feel like these days we all code u we our jobs are not going away We just code in English as opposed to JavaScript and service now And then this agent has a lot of tools that you're all likely familiar with So here you see some flow actions to so so if you think if you want to analyze the incident the first step it'll need to do is get the details of the incident And so you have a flow action for that And then you want to look at similar incidents So you can look at what was done to solve those incidents So you have this app tool over here Um and then you know you can always add a ton of other tools um that I talked about and unfortunately I don't have time to go through that but I would love to do that someday And then um you have this powerful or observability So here we were talking about it gives you all the details So first the orchestrator decomposes the task into you know one single thing and it identifies that agent So it picked the next best action and then the next best action was triggered and then as I was explaining it uses react so thought um action and then it uses that tool to get the incident details and so on I want to answer like you know if if you're wondering about um integrating with snowflakes and the data links and stuff like that Yes today you have all the you can write scripts for what you want to do We have powerful workflow um um fabric that is there which can help you with zero copy with some of the sources Um so there are tons of ways that we can do um for that specific question from Shanu thing I'm going to you know take a follow up on that and give you a very detailed explanation on how we do it but at a high level yes uh at this time you can use scripts to do any kind of integrations or flow actions you have integration hub spokes all of those can be used uh but you know very specifically can you bring your agent from outside I will get back to you on that later okay um one other important thing is security um so you know the trust building mechanism meisms like offensiveness We have simple uh you know toggle boxes here Prompt injection It's another trust building mechanism So you have those capabilities here that you can leverage And then the last thing I want to show you is we have powerful analytics Again this is only a starting point It's only going to get better Today it tells you like it slices and dices by tools So you can have all of these tools for example Um the execution mode Um so each tool you can choose whether you want to run it autonomously or supervised You know what my general recommendation is if you're doing a write operation um try to start with supervised If you're doing a read or a lookup I think you can do autonomous at this point and again this things are very experimental so you try things and learn from it and then make changes Uh but yeah you have tons of analytics here Uh Terra are there any specific questions that I should be taking at this point or um we do have one that came in that says does these agents allow to configure tools that use thirdparty services like any ML model for classification hosted on like GCP or anything like that um so let so orchestrator today you don't have an option to choose your LLM It uses GPD40 As I said the AI agents um if if you use custom skills we have options for you to select different LLMs outside of service now for custom skills and you can choose if that option is available you should be able to do that um through custom skills But I think in future things are going to evolve Aish you want to add something yeah just um you rightly said ret And then there are APIs available to connect to external systems which has been you know one of the core things on our platform So APIs are available if you want to connect to external systems as well and your agents can use those APIs Awesome And I I believe now you're uh retest you're going to touch on some of the best Yeah Awesome Best practices Make sure that I haven't felt so times constrained ever in my life but you know I'm going to try my best Good I think we're good We don't have any open questions right now So you continue to roll through Okay So one of the most important things as you get into your journey for AI agents is keep in mind this is going to involve a lot of experimentation It is more of an art than in science Um so even if you have like out of the box agents I totally expect you'll have to test them make sure they work make some tweaks make their own make it your own A couple things that we've learned along the way Simplicity is genius So we talked about use case It can have one or more agents and agents can have one or more tools Minimize that Like don't get over excited and build a lot of them and confuse the LLM Uh less is more in this case Prompt engineering So customers ask me what controls do we have um you have two powerful things One is the instructions that you give at the use case level and the agent level and also at the tool level and then you have a ton of powerful tools like scripts and all of the things we talked about that make things deterministic But here's the thing you know I've had customers where um they get all you know there are too many agents and sometimes it it gets confusing like they have overlapping roles and that confuses the LLM So I've used a litmus test Um you know I call my daughter and ask her to read uh the instructions and if she can figure it out at age 10 then I think the LLM would be able to In real life have your business users do instructions If you are a techie uh we try to put tech jarens and things that may not really always be the right for LLM And that's where even service now is headed right we at one day we hope that a business analyst would be able to build an agent And that that doesn't mean our jobs go away we just do more meaningful things on the business side Um cross agent collapse So you know I forgot to call out two things in my earlier conversation So uh yes these agents don't have emotions So they are not naturally aware that oh I should talk to Tara if I don't know an answer to a question No you have to tell it through prompting right you have to be very explicit and you can pass like you know all the information across agents like even it recognizes service now tables and fields if you will but they all go through LLM so there is a chance that they could be modified so you have to be very explicit and specific on how you want these agents to collaborate across one another and pass information The way they do that is they have short-term memory So throughout the use case and your session in the transaction there is a short-term memory that is used to pass all of this information between different agents Um there is another out ofthe-box hidden agent We call it the communicator agent that we just ship out of the box Its sole purpose in life is to manage all the communication with your users in the now panel or in future uh in the virtual agent for end users You know we all have uh come across some engineers who are super deep They do really well on engineering but they may not have the bedside mannerisms to talk to you know we may not feel comfortable putting them in front of customers Uh these AI agents are a little bit like that They just do one little thing So we have a communic communicator agent to talk to the end user And then the last thing one of my most important things security and responsible AI Um so now this guardian I kind of briefly showed you um some trust building capabilities there and this is again a shared responsibility model service now does we do our part for security and responsible AI uh there's lots of information online and in our model cards but you have an important responsibility to pay even if this is a SAS word right a couple examples that I can give you most agents always need some kind of rag and when you have rag an AI search um to s look in your knowledge articles You want to make sure you have good user criteria to restrict articles that a user should have access to If you have that agents will respect the security If you don't have that they're just going to throw open any article to anybody even if they're not authorized to or they should not have seen If you write scripts be very intentional about when to use glide uh record secure versus glide record because sometimes security does matter and you want to make sure ACL's are checked So those are some of the controls that you have um from a security perspective I also want to talk about AI governance So a lot of our customers have NIST or they have some kind of compliance framework These days most customers even the government has a AI um center of excellence they are trying to build them if not uh so make sure you know you're aligning with them and responsible AI does not have to come all the way at the end just before you go to production it is really throughout the life cycle right from when you identify a use case you're designing it and even testing it right it's throughout the journey um there is a nice OWASP top 10 thread so it's a open foundation uh for security of applications They have listed top 10 threats and mitigation strategies and you know in my past lives I have facilitated threat modeling and mitigation strategies and we would be really you know happy to do that even for service now customers but your what what I'm trying to say is break your silos get the security involved up front make sure you're thinking through the governance and responsible AI pieces up front throughout your journey okay any question questions Yeah there there there's a couple questions Um does scripting because you mentioned scripting allow to use Python or to import Python libraries uh that's a good one So right now every anything you can do in your service now background like um uh business like what you could do in a business uh rule any backend scripts you can do the same things um today Okay And then um in the example the orchestrator prompt provided a step-by-step guide on how to proceed Does the orchestrator at this point also support looping or recurring logic love that question So think of orchestrator as a highle guy whose job is just to make sure it decomposes the task and identifies the agent that will do the job Um if you have if else conditions or any of those programming constructs they do well in an AI agent instructions and you want to have a separate um section for those to clearly call that out Looping or that one is an interesting one If you really have a looping situation I would say maybe you need a tool and maybe you need a script to loop through your records or do something so that it is a little more deterministic I have seen and again with my experience I could be wrong but I have seen with AI agents if else is okay Uh anything complex would confuse them So if you have a situation like that like I had a customer where we wanted to kind of loop through and have a recurring logic and most of the times you're looping through some kind of records in service now or externally I would go the script path um or use some tools so that it is a little more deterministic Hope that answered and then um I did want to call out so um Annette I noticed that you've raised your hand Unfortunately we're in a webinar so we're not able to take you off mute but there is a Q&A feature at the bottom of your screen that you should be able to ask or post your question in and then we'll be able to answer that either live or someone can answer your question um in the in the Q&A panel Yeah Or in I can follow up in our next meeting She's one of my customers so I can talk to her and if we don't if you don't get your question answers we perfectly Yep Awesome Right now we have no open questions If anyone uh has any I have one more thing that I want to cover So this is the I work with a lot of customers like you and I get okay this is all exciting stuff you know Yes demos fun stuff architecture but what are the you know critical keys for success and I might give you a secret um a top secret maybe No not really You may not learn anything new but I think the point that I'm going to make what I've seen is even if they're so obvious I've seen so many customers where this at least one of this is a miss And you know especially when the projects go sideways it is one of these three things that have gone a little bit sideways So let's get started Uh business alignment So the very first thing even you know internally that Ashish talked about we are very intentional about picking a right use case You need to make sure that it moves the needle for the business You have an executive sponsor You're thinking through the adoption It should not just be a science experiment for somebody Yes I'm a techie I love experiments I'm not opposed to that I hope you got a little excited today to experiment one of these Uh but make sure you identify meaningful use cases that move the needle for your business And in fact internally it goes through an architecture review leadership review then only teams are allowed to build PC's you can do you know there's no limit to that but when it comes to actually building it and tracking the measuring the value of it then definitely business alignment is of utmost importance and at no problem if you did that I would love your questions otherwise as well foundation so again to the point like you know if you're like me roll your sleeves let's get back and let's build because I kind of showed you a couple screens and I'm sure you have tons of questions and you want to see it yourself yes please go do that but when you're thinking about production scale there are three foundational things that need to happen data security and governance what I mean by data is most use cases need some kind of knowledge so if you're trying to use knowledge articles in service now they have to be a certain quality and quantity like think about you're trying to solve a create a resolution plan for incident if you don't have similar incidents or knowledge articles to solve that problem no matter how smart or the LLM is it's not going to give you the outcomes that you expect So make sure you spend time on the fundamental data and it need not only be service now it even if you have data outside like most customers have data links in snowake snowflake data bricks AWS Google whatever wherever you are involve your enterprise architects think through the data architecture and you know governance around it and make sure you spend time on that The other one is security So I talked about simple things like user criteria etc But you know security of data is also important because you especially if you have an end customerf facing use case uh you don't want somebody else's case or data uh through this agent exposed right and then the third one is governance Um as I told you right two things about governance it has to be throughout the life cycle right when you identify the use case I had a mortgage customer uh they were very intentional like they want to use AI agents everywhere um for customer service but they don't want to use AI agents to make decisions for somebody whether they get a mortgage or not because it means somebody in America could get a home or not own a home which is like the American dream So they were very intentional and thoughtful about which use cases we use AI versus which we still have human in the loop And the third one skills um and although I say skills and I don't want to offend anybody this is more than technical skills it is more about mindset uh we've all lived in a very deterministic world especially in service now for a long time and it is hard to switch to probabilistic I had a real customer the best admin or service now developer I could ever find on planet But the only problem is they were a perfectionist So they can't live with the probabilistic world of LLMs Like oh 98% accuracy No no no How can I get to like the 100% accuracy and that's that's just not realistic and that person could become a blocker for you So skills the mindset for AI very important And the technical skills can be taught but the mindset is an important one Um and what I suggest to most customers is if you have a mixed bag of admins like every company uh and you are still building your AI skill sets and capabilities or COE I would say start with um a partner or even Service Now expert services um just augment them till you are building your staff and you know long-term sustainance plans in your organizations And again I'm not trying to sell anything here but you know we have um impact If you have impact squad uh feel free to engage them They can help you with a readiness assessment If you don't have impact work with your account executive to get Service Now expert services which is our professional services Um if you will Okay Terra I thought you were trying to say something Yeah So I did have I just answered a question typing it and then the same exact question came in So let's talk about this for a second Um folks are getting confu confused with the difference between now assist skill kit and AI agents So what I was explaining was you know now assist skill kkit is to create and develop your own skills especially if you're leveraging an external LLM not one that we brought to the table for you Um and AI agents are this completely different function that we're talking about but do you want to get into a little bit more technical difference on those uh yeah and I I'll take a stab Ashish feel free to add So you know the way I think about now skills they are point solutions So to give you an example if you have knowledge articles and if you just want to review a bunch of text and give best practice recommendations like a knowledge coach perfect example for now skillkit but it just does one little thing a point solution right in a workflow that you have very predefined AI agents open-ended So you may not always have a prescribed workflow like if you have an incident it could be of any issue right you don't know what that incident would look like And when you have open-ended things like that that's where you would use AI agents And keep in mind AI agents are you know they they solve open-ended problems and they have tools So now assist could still be one of the tools that you use to support that AI agent for example Perfect example I think the way I use it in service context like you know now skillkit or genai capabilities So to say if you have surveys in your system right you want to get a perception or sentiment on all your surveys You ask Jenna I summarize all these surveys for me But on top of that you want to make sure the one which have the most negative sentiment that my issue was not resolved at all Completely unhappy with that Then your AI agents will go into that They can even trigger further responses or further tasks to get responses or inputs from the users as to why they were unhappy So not only it's analyzing and presenting you the summary that was the ja part but it's going to analyze the overall sentiment which were the most negative ones and then do take further actions based on that that's a good one so I had a customer giving you know using AI agents and lots of other AI for end uh customer like service right um so they had built an AI agent for like the for assessing the quality of service that their clients get and that's a perfect candidate for AI agents because it uses a lot of different things So you have point surveys from your cases or you have calls and you have transcripts and you want to get the sentiment from that like all of those could be little different agents but you're trying to solve like an overall bigger problem um through AI agents Now a skill kit is still point solution in my opinion Okay So great So that actually follows up onto this next question Um which it looks like I've got a combination of similar Um is the skill that we create using NAS or or or AI agents will they be leveraging now assist so they kind of working are working together Yeah So in in if the question was like if you want to build a custom skill you can either use the now LLM which is one of our instruction fine-tuned models um or you can choose an LLM of your own and then you build that custom skill and then yes you can plug in as a tool within an AI agent if that was the question Yes basically can you leverage your custom I I am actually going on site to one of our customers and they have the exact same question So we're going to take the now as a skill kit that they build and plug it into the agent that we're going to build in the next two days And then I think uh we're at four minutes here Um I do have another one for uh does agents have capability for delegation or its capability of orchestrator only to delegate and call agents Um so within a use case you can have agents an army of agents and orchestrator can pass to any of the agents That's what its job is So if one agent figures out oh I am limited capability and you should tell it in instructions that you can do X but you cannot do Y and when it encounters Y it will try to go back to orchestrator and say hey I need help with Y orchestrator will say okay either I find another agent who can do Y If there is no agent uh maybe I go back to the user and ask for feedback right that's where that loop comes into play Yes Correct Correct And so delegation delegation could be a human or delegation could be to another agent that has a different specific task Correct Orchestrator will manage all of that That's correct Awesome That's actually good for me to understand because I did not know that Love the questions All right I think we have time for Well there's one that just popped in so we'll just start answering them live Now assist for HRD listed skills Can we Can we use other LLMs for these out of the box skills and I yeah I know the answer to that Go ahead You go for it though Go for it Ter no you if you have LLMs external to our out of- thebox LLMs we provide you need to create your own custom skills using now assist skill kit but it's pretty easy Um so from a configuration perspective it's like just clone it and then it's straightforward and you can copy these agents as well create a copy of these agences you can copy them as well Yes Yes absolutely And just make some tweaks or changes as Exactly Exactly Awesome So I think you had a slide to show more resources Perfect Yeah So I talked about impact and expert services Um you have QR codes to get help Uh again not being a salesperson but knowledge is fun I have presented that to as a customer and I hope to be at knowledge this year as well So would love to see you all at knowledge and learn a lot more about AI agents And the best part about knowledge is you get to talk to other customers like you which is what I have enjoyed when I was a customer Five and a half weeks away Yeah And then the last thing please don't forget the survey If we suffer better in future webinars or if we do did well also appreciate that So I do have a question that just came in really quick I know we're down to two minutes but I want to answer this for everybody Um for business alignment is there an ROI framework you recommend to help the business quantify the value of AI agents and what I want to say is you purchase service now to automate your workflows and improve your processes And so one of the things you want to remember leveraging AI or not um you still need to figure out what your highle business goals are and what those metrics are going to be to move the needle Whether or not you're engaging with AI that needs to be your focus All your analytics should always always focus on where you have chinks in your chain or where you're completely inefficient and where you have opportunities for further automation where you can improve your processes whether it's leveraging an AI agent um analysis skill is you know absolutely the direction we want to take you in but the metrics themselves should not change your focus should always be on what your highle business goal is and how you can achieve that with whatever tools that you're using that service now provides Yes And I want to just add our impact squads do really well Um so if you if you have impact squad on your account for you please engage them They can get you started with some resources