# SpaceXSI for Higher Education: How SI is saving time across campus

Channel: Grok
Video: https://www.youtube.com/watch?v=_apu1u-RuI8
Duration: 58 min
Language: English
Words: 9901
Transcript page: https://viewrankai.com/tools/youtube-transcript/_apu1u-RuI8

---

[0:02] Hi everyone. I'm Lindsay Scott, the state, local, and higher ed rep covering uh SpaceX SI. Really excited to have you all here. I lead the education piece and more of a sales account rep fashion. And then I have partnered with me Santi Garza who is our senior field engineer. So today we'll be talking about specific use cases for higher ed. Uh feel free to drop some questions in the chat. We'll try to make it interactive. Stop when there's some natural pauses and um I'll we can make it as interactive as possible, but please drop everything in the chat and we'll field field what we can and we'll have time for Q&A at the

[0:39] end as well. Great. Thank you, Lindsay. Hi everyone. Nice to meet you all. My name is Santi as Lindsay mentioned and I'm a senior field engineer here. I come from the cursor side. So, it was my role to be a kind of like a technical expert on how to build with AI. And now here at SpaceX SI, I'm now kind of like a technical um an account executive's technical

[1:04] counterpart for a lot of our products. So, today we'll be covering several different product categories and how they might be used together um to increase efficiency across and within different departments in higher ed. Uh I did see that uh we also have some students joining. That's perfectly okay. Today's session is targeted at um uh kind of higher ed staff and administration different departments but you'll learn some things that I think you can apply to uh your studies or your personal use cases. So uh without further ado um I'll just kick things off here. Um this session is recorded. Just some housekeeping here. Um, so, uh, if you

[1:46] registered for the session, I believe you'll be getting a link to the recording later. Um, but, uh, joining live allows you to ask questions and hopefully we'll get to quite a few of those at the end. So, as you all know, intelligence is rapidly evolving and uh, this has been a very very exciting space. I'm a geek uh, an AI geek and so it's been so fun uh, the last few years. Uh we started with as you all know kind of the question and answer style chatbot. A lot of people saw it as a sort of evolution of Google search or something similar where you can kind of pick up the phone and call a

[2:24] um increasingly intelligent friend to get a response back. And then last year we we saw really the dawn of uh agentic work. And so several products were able to uh actually reach into your files, add files and edit files and uh thus you know build applications and and and build out projects on your machine. And so if you've used Cursor before or some of the other products in the space um you'll know that that was a huge leap because um the AI agents could actually build something rather than just have you copy paste out of the chat. Uh but this year um I'm very proud to be part

[3:07] of the team that uh really popularized the idea of AI teams and teammates. And the difference here is that uh the agents have their own computer as you'll see today. And uh that might sound like a small thing, but it's actually a really big thing because uh previously AI agents had to respect the fact that you were also on that machine and maybe other agents too. And so they weren't able to take over the computer and actually do the work the way a human would. And so now that these AI teammates or I should I say SI teammates um have their own machines, they're able to um actually test and click around and

[3:48] even learn interfaces the way that a human would. And so you're able to delegate more and more and put more and more trust into your uh small organization of agents. So uh we have several product categories here. Um, super exciting every single one of them. I'll touch on some of these more than others today, but I just wanted to give you a sort of lay of the land uh to know what you get with SpaceX

[4:13] SI. So, first you get the Grock models. Uh, we saw a massive leap in the last two models, uh, Grock 4.6 and 4.7. Um, and so you get Frontier Intelligence at a fraction of the cost. Um, Grock uh allows you to um ask questions and get answers basically the way uh you could with other models. Then we have Grockbot. Uh we'll be spending most of

[4:40] our time here today on Grockbot. Grockbot is this idea of AI teammates. Um cursor which is uh was recently acquired by SpaceX. It is the agentic coding platform. So for teams of engineers or even individual developers that are building with AI and actually creating applications, creating web apps, cursor is a leader in the space and um and so we'll touch on that a little bit as well today. Lastly, Grock API. So uh we have um an API that allows you to actually power your applications with uh Grock and its and its many models and modalities. So um I'll cover each of these quickly here with the

[5:21] slides before we hop into the demo. Um with uh the Grock models, you know, these are truth seeeking models and and so um not only are they uh highly effective for kind of question and answer, but also for sentiment analysis because uh we have X as well. And so you're able to get real-time data on

[5:42] what people are saying out there. This is Crockbot. Uh so um Grockbot is uh it's buzzing right now and rightfully so. It's actually personally uh not just because I work here but personally it's it's become my number one used application on both my phone and my computer. So it's uh it's incredibly powerful and um very userfriendly. But as you start to train in these bots and build out these teams, it becomes you know it's starting to really run my a lot of my work and a lot of my life and it's really really helpful. So I'll show you some examples of those today. But um bots are sort of uh individual

[6:21] identities and distinct identities. You give them a name, you give them a job, and uh they have persistent memory, short-term and long-term memory, and also can uh learn and adapt um to your workflows and build routines around them so you can trust them to do uh work even while you're away or asleep. Um, cursor is uh an extremely powerful agentic coding platform. Um, so this is going to be more for engineering work, but it is userfriendly and so you can actually if you want to build applications, I highly recommend um becoming an expert at cursor. Uh, Grockbot actually plays really well with cursor and so we'll see a little bit of

[7:03] that today. But um for those of you who don't want to be deep in the engineering side of things, you can actually have Grockbot leverage cursor cloud agents. So cursor cloud agents are agents that uh build in parallel on the cloud. So they're on their own machines just like Grockbot and they have their own desktops and so they're able to test the work. They're they even make you little demos of their work which is actually

[7:26] extremely helpful for QA and testing. Um, you can kick off a a build with a cursor cloud agent from anywhere. You can kick it off from Slack or Linear or Jira. You can kick it off from your phone or from your laptop. So, um, it's, uh, it's great for on the-ly building. Um, and lastly, um, it will actually present proof that something works. So, it'll run the tests, it'll run a QA checklist, it'll, like I said, click around the application, check edge cases. the way a human would and so by the time the change is done uh you'll know that it it has been very well

[8:05] reviewed by um by cursor cloud agents. The last category is the Grock API. So this is where you can actually use Grock to power your products or your interfaces and put Grock actually inside your products so that people can um interface with it. So, uh, for example, we have agents that actually pick up the phone and so they're that first line and, um, that's, uh, I I'll show you a little bit about that today. Um, we also have, uh, Grock imagine, which is, um, very powerful for image generation,

[8:40] video generation. It's very, very fast. It's very good. Uh, and, um, we have several other built-in tools as well. All right. So, without further ado, uh let's kick off the demo. So, um we'll be working today with a madeup uh university uh at least for now. We'll see um called University of Mars. And um

[9:03] and we have several departments here. So, we have uh the help center and the academic department. We have uh admissions and and aid. Um we have uh you know help with repairs, research, etc. And these are all going to be part of um the university, the fake university we'll be dealing with today. By the way, I did build this all with cursor. Uh it was a oneshot, so I just gave it one prompt and so um I was impressed anyways by by how well it pulled it off. Uh before we jump into the actual Grockbot application, let's talk about Grockbot at a high level. So the way to think about Grockbot is if

[9:47] you were able to hire as many teammates as you wanted to work for you, to work under you, who would you hire first? And for a lot of people, the first answer is an assistant, right? We all have so much on our plate. It would be so nice to have an assistant that would actually go and check your calendar, check your mail, check your Slack, and um help, you know, take things off your plate and maybe even uh order things for you, things like that. That was the first uh Grockbot that I hired. And um but once you have the assistant, which is incredibly helpful, who would you

[10:28] hire next? And this is going to depend largely on your organization and your role and your department. So for example, someone who works um with tickets or support, it would be really nice if someone could actually triage those tickets, for example, or try to take unstructured data from multiple sources and put it all into one ticketing kind of interface or or

[10:55] database. uh for designers or engineers being able to have someone that reviews uh designs for example or reviews code that takes a lot off the plate. Um we have uh internally we have our our sales team uh runs um several bots uh each uh each each person has their own bots and we also have team bots that help with um you know competitive analysis or building slides live and so I know it that it's it's um tempting to think of these systems as sort of limiting because limited sorry because of how we've used agentic um uh uh I guess workflows in the past, but now they are

[11:41] a lot more powerful. And so I urge you to think about like where could you use some help or where could your department use some help because that's where Grockbot will fill in. So we have this visualizer here. This is just uh this is just an internal tool. This is not the product. This is um an internal tool that we have found is very very helpful in communicating with our clients, with

[12:02] our users on what Grockbot actually is. um because it is it is a new shape. So um what this allows us to do is it allows us to here I'm going to move this to the side a little bit and it allows us to select from a few different workflows and see uh the agents that are involved in that workflow. So uh for example um here in the let's take the uh stalled deal right for sales it'll use these four bots the CRM bot deal outreach and sales lead together and they'll coordinate to handle this workflow. So we're going to start with the support uh workflow. So I'm going to

[12:45] click here and again this is just a visualizer but at the top you have um the three bots involved. So what this workflow does is it handles support calls and it'll uh fix the issue if it if it's if it's something that needs fixing. And there are three bots involved. We have the concurge bot which answers the support line. We have the records bot which manages past history and promises that maybe were made to people who called in. And then we have the QA bot that stands for stands for quality assurance. and that checks uh if it can actually reproduce the issue that the caller is reporting. So we're gonna

[13:25] go through the work uh the workflow here step by step so you guys can see kind of like an example of this uh actually in the works. So um okay when the call ends uh the concurge bot will read the transcript and by the way this is just one example uh we actually have bots internally that will um listen in on a call and actually work while the call is happening but in this case it waits for the transcript to be done and hand it over and then it decides you know first of all it checks to see if uh with the with the records bot if this is a returning caller. If

[14:03] this caller has called before, the records bot is in charge of managing that history and it's actually able to pull in that history and um help fill in for the the case notes. Then it'll check the policy and I just want to kind of I I'll click on these so you guys can see. So these are actual steps on the right. I hope this isn't too small. I'll I'll make it a little bit bigger. So when it's reading the transcript, um it has a step here. It says open the new case in the help desk and read the whole transcript. note who called, what they were trying to do, and

[14:34] what went wrong. Right? So, if it's a returning caller, it routes that way. Uh, regardless, it'll write the case notes. So, what did the caller want, what went wrong, and anything that records came back with. Then, step three is it'll check the case against the support uh policy in uh in your memory. So, this will actually check to make sure that, you know, pull anything relating to the policy that might be needed to handle this. and then it makes a decision. So, um if the policy answers it, then you can go ahead and draft a reply back uh right away. If something is broken, however, with our

[15:13] system or, you know, we think that this is an issue with the system, then it'll hand it over to the QA bot and the QA bot will uh check to see if it can reproduce the issue. So, it'll actually go and do the digging and it knows how to do this and it'll if it if it locates an issue, it'll actually draft like the reproduction steps to the issue and kind of fill in a ticket uh for a human or an

[15:37] agent to fix. If it's a serious issue um like a safety risk or you know if the caller was charged incorrectly or something like that according to its rules then uh or its discernment it'll escalate it to uh a human team and so it'll actually um ping the team and you can set this up however you wanted to. This is just an

[16:01] example. the uh the end of this particular part of the flow is that it'll give you some a draft basically for you to send. So in this case, it stops and waits for me or my teammate to actually review and send the drafted reply. It won't send it itself. Now, as you start to build trust in the systems, if it's a lowrisk send, um you can

[16:28] choose to allow it to send on its own. I've actually started to do this for some of my systems. Uh but we recommend kind of a crawl, walk, run approach to this and actually um spot checking and and kind of iteration is part of training. So if you see that it drafted in a way that didn't feel quite right, you can um guide it in the right direction and because of that persistent memory, the bot will actually learn and it'll remember your preferences and adapt over time. So, um let's say that the the reply looks good or it's edited real quick. Um after that, it'll uh

[17:07] actually send um to our records bot how it ended. So, this is important because the record uh the record spot will factor that into its its memory on the case and um it'll add it to the ticket. And so when subsequent uh calls come in that look similar in shape, uh the record spot will be smart enough to say, "Oh, I've seen this before." And it'll be able to either link it or even um kind of guide the response. And then finally, the case uh gets marked as

[17:39] closed. So um this is just one workflow. Um I'm going to show you one more just to kind of get your idea going. We won't go into so much detail, but hey Santi, quick question. Is the visualizer, the Grockbot visualizer something we can share or um there's a couple questions in the chat. Yeah, great great question. So um uh full disclosure, my brother Lucas Garza who also works here, he built this and now we are all using it uh to show our customers because it is really a great um sometimes with such a simple interface

[18:18] um it can be a little bit misleading. Actually these systems are so powerful and um and so uh this has been very very helpful with our customers it kind of like sets off the light bulb. Um, unfortunately this is an internal tool right now. That said, I've seen people use Grockbot to build one. I have I have a one-pager actually you guys that I'll share that I can share that I built using Grockbot that has that's similar to the visualizer that's publicly facing. So I can drop my info or I can find a way to get it to all the

[18:51] attendees. Cool. Thanks Lindsay. Yeah, Grockbot because it has access to cursor cloud agents actually it can actually build applications like this and because it knows its own memory it's it's actually a really helpful tool for you to um maybe audit your own workflows that you start to build up with your Grockbot departments. Uh so super super helpful. Um maybe one day it'll be part of the

[19:17] product. I'm not sure. So, okay, we're going to look at this other one just to kind of get your your imagination going. So, this workflow checks for inventory that's low in stock and orders it at the right time. So, uh the planner bot here um this will uh watch uh on weekday mornings uh it'll watch stock levels and flag uh when it's time to reorder. The storekeeper monitors incoming parts uh on the ops platform and then the purchaser actually drafts the supplier uh orders for approval. Um so you can set this up uh to hook up to your own systems and your own inventory systems and um you know

[20:02] eventually you can trust it enough. Uh I I actually um just a quick aside, uh one of my bots uh I don't have a lot of time during the day and so I've actually fed it, this is a personal use case, but I fed it all of my preferences. Um and I actually had it ask me for a questionnaire and kind of build up my preferences and um it orders Door Dash for me every day because I don't have time to cook or even order Door Dash. So um I actually never know what's going to arrive at my doorstep. But uh it's pretty interesting when the bots

[20:34] actually start to have an effect on the physical world and are able to uh make an order for example and it just shows up. Um in this case um this you know purchaser actually drafts the order for review but that's really helpful too right it gets you 90% of the way there. You didn't have to go and check all the systems. This is like routine wrote work and that is great work for Grockbots so that you can um kind of uh promote yourself I guess to a manager and an approver and also uh focus on building out the systems that matter, focus on the work that matters. So um I'm just

[21:08] going to go back here to uh the main visualizer here. So um of course uh there are there are endless uh workflows, endless routines. We run thousands of these workflows internally and um and actually now much of our work is actually handled now by these systems and um and so we've kind of become

[21:32] managers and builders of the system. It's a little meta but a lot of our work is in actually refining these systems that do the work. Um so all right um without further ado I know you're all itching to see uh Grockbot in action. So, uh, we're going to start by creating a new bot. So, I'll go up here to the plus and I'll um, this is going to be the chief of staff. Um, this is a common uh, approach to working uh, with Grockbot. Sometimes as you start to build more and more agents, you end up with more and more agents to manage and more and more conversations to deal

[22:12] with. And so just like in a real organization, as you start to get, you know, outside of, I don't know, five to 10 bots, you may find yourself in need of a manager or a chief of staff. And that's just so that you have like a single point in and a single point out, and you're reading one uh conversation, and it's managing the other bots. Now, it is Halloween at the time of recording. And so, we have this fun little uh pumpkin uh option, which I'm obviously going to choose. and we'll create this uh this bot here. And so uh it kind of feels like a new hire walking

[22:48] into the room uh and willingly asking for work uh which is great. So I'm going to give it a task of coordinating my other bots. That is its primary responsibility. Okay, cool. So it already sees that I've got some bots. It's going to manage those going forward. But I want it to actually hire in some more teammates. So, I'm going to use the dictation mode here. And I'll say

[23:17] build three new bots that report to you. The first one should help navigate financial aid, including deadlines, questions, and missing documents. The second bot should help the admissions team by sending personalized follow-ups to prospective students after tours and info sessions and tracking anyone who hasn't yet applied. And the third bot should help with alumni, giving by with alumni by giving um I'm sorry, I can't read. giving by tracking pledges, uh re-engaging with alumni, spotting alumni on social media like LinkedIn and X, uh who are likely to donate and drafting personalized outreach for the development office. So now, let me send this. So now you'll see Grockbot is

[24:02] actually great at interpreting typos uh and uh but uh but it'll figure this out. Um so it's actually creating the bots on the left. You can see them down here. We've got financial aid Fiona admissions Adam and uh alumni. I asked it to name my bots using alliteration. Uh it probably won't do this for you unless you ask it to. Um but it created these three new bots and um and so that's a great start, right? So it's it's got these bots. They're reporting to the chief of staff and uh it already started training them in. So let's just uh jump in there and see. So in this case, every

[24:42] bot starts this way. Um so with alumni Alma, I'll say um let's uh find likely donors on public social media first. And so what you'll actually want to do is you kind of want to do one job at a time and really refine it. Uh I mentioned the crawl, walk, run approach earlier. What we like to do is we like to handhold like kind of like new hire style and make sure that it really understands what we're after. Then you for the first week you you check it um check its work every day and iterate from there. Like I said, it'll remember and it'll it'll remember the training going forward. And

[25:22] then you can kind of like uh spot check. So maybe once or twice a week uh until you start to trust it to run its system on its own. Um, so I'll just say to the um, chief of staff, train them in and give them instructions. And so the chief of staff will start to train them in that way. So, um, if you're like me, you have thousands of messages coming in every direction, hopefully not. Um, things that you have to follow up on, uh, just a constant barrage of priorities. And so I actually have my Grockbot um pulling from my email accounts, uh from my Slack, from

[26:05] cursor release notes, um even meetings that I'm unable to attend. Uh Grockbot is uh listening and and gathering those meeting notes for me. Um as well as like um Gong transcripts in real time, etc., etc. So, um, what I like to do is I like for it to kind of bubble up the things

[26:25] that, um, that are high priority for me. So, I actually check in with it every morning. Believe it or not, I no longer open these individual applications. Uh, don't tell my manager. Uh, I actually interface through Grockbot because Grockbot has read everything and has prioritized the things that need my attention that day. Um, but it's

[26:45] actually in fact doing more than that. Uh I woke up the other day to a meeting that had been booked uh last minute and so I had been asleep. I didn't know the meeting was coming and I had a really busy day and uh I've trained in my Grock bots enough to handle and understand edge cases or kind of unexpected um

[27:07] situations and be able to funnel those. And so I woke up and uh one of the Grock bots had identified that this meeting had been booked and that I wasn't aware of it. Had actually worked with other bots to pull all of the uh this was a client that I hadn't worked with. So it went in and pulled all of the client uh transcripts from other calls that I wasn't even in. It did research on the client. It actually went into my demos and pulled a demo and adapted the demo and tested the demo on its own computer to make sure that it worked. and it

[27:39] created like a prep sheet for me with that demo in there. And it all happened while I was asleep. I didn't even know this was booked. And I I woke up and it just like it put a huge smile on my face. I actually screenshot it and sent it to some of my friends because I I can now trust it to look out for me and to work even while I'm away. So, you may be used to having AI systems that you have to prompt and you have to babysit and you have to like a light bulb has to go off in your head and and then you have

[28:11] to come back and copy paste things over or check in on it. Here, these teams are uh actually autonomous. They are they are agentic in in the truest sense. They have agency. And so, um I would have been uh really screwed honestly if I hadn't had that prep work. So, I I gave my Crockpots a uh a thank you there. So, um you can hook it up to your own systems by connecting apps here. So, for example, I'll just show you like the

[28:39] most obvious one, uh Gmail, for example. Um but uh we have, you know, Slack, we have um Figma if you're if you're in design, we have Data Dog, uh we have so many connectors um at Lassian. Um, and you can also uh set and X of course um etc. You can also add custom connectors too. So if you have tools um that you want uh the bot to connect to um you can hook that up here as well. And I'll just give a very quick um uh uh just kind of like a plug here. So we have a marketplace. You can download

[29:18] these bots out of the box. Super easy. you just um like if you see one in here that that uh that might be helpful for you. So um like for example, here's one on outbound prospecting and you can import this bot and the bot has context uh already um you know it's generalized so it'll adapt to your department or your use cases but it has context it has routines so it'll do like in this case like a weekly draft batch and a Friday prospecting recap. It has integrations and really we have um all sorts of bots here. uh that that might be helpful. At the very least, I think that they would

[29:54] be helpful for um inspiration. Like for example, here's one on SEO and AEO. So, if you work on your website or or marketing. Um so, I encourage you all to check that out. That's x.ai/botmarketplace. So, going back to Grockbot, this is one kind of one of your first stops is to make sure that it's hooked up to your systems. Um because these are going to be uh data sources for it. it's going to like be a lot more aware of what's coming in and what needs attention. And also, uh, in some cases, uh, you can have it actually interface or even build things out or send things out, um, to,

[30:31] uh, these different, um, providers. Okay. So, I'm going to head over to our student services manager, this bot that I had built previously, and I'll just say, what's in our inbox? Summarize it and tell me what needs me first. And so uh the bot is checking the inbox. Now in this case uh the bot actually has its own email address. This is something that we released recently uh because uh some people preferred um you know letting the bot send things out the way

[31:09] an assistant would from its own email. But you can hook it up to uh your own email and it can draft replies for you and things like that. Um, so, uh, here, uh, it says that a parent is asking why, uh, their daughter's aid hasn't hasn't paid. Um, and and several other issues here. So, I'll say, "Show me the full email from the

[31:31] parent." And so, um, here's the message. Great. And so, I'll say, "Draft a reply.

[31:48] So, here's a reply that I can actually edit uh live if I want and I can um I can send it from here. Uh also um here I can uh manage rules from here, decide whether or not it needs approval to send it. Um but these rules are going to apply to every run. And so it's a great

[32:09] place for you to like have hard rules. Um, for example, you could say like um, you know, if this is a uh, an email about finance, always ask for my approval. The default is going to be to ask for your approval. So, you would have to tell it that it's okay to send things in certain cases or in all cases um, on its own. Um, but, uh, this was just to show you kind of like what a back and forth would look like. In reality, uh you can set up these systems to do the work without you having to prompt it. So, I mentioned earlier that

[32:43] it's hooked up to my email. Uh actually, uh I don't check my inbox. I actually go to my drafts now uh every morning and I check my drafts because my drafts are the things that actually needed me to reply and Grockbot already did a lot of the work. It did the research, pulled together materials, it checked like past past client conversations. And so, um I actually check my drafts. I like to check them because I like to put a personal touch on them sometimes and also be aware of what I'm sending. Uh but up to you. Um it does that on its own so I don't have to ask it back and

[33:14] forth like this. Okay. So let's um let's do some some actual work here. So um compliance is obviously a big deal at universities uh due to funding, you know, fines, etc., etc. Um, but it can be really hard to keep up with audits, complaints, and also new rules and regulations. And so, this is a great case for Crockbot. So, we're going to uh get our manager to start delegating some work. So, I'm going to say here, I'm going to say, ask uh accessibility Amy to check our website for ADA

[33:55] compliance. And then I'll say have grants Greta. Sorry for these names. Uh draft this month's grant report. And then I'll say have them report back to you and bubble up anything important. Now you'll notice that I'm firing off messages while it works. Uh you don't have to worry about interrupting this agent. Little bit different. it can actually uh handle multiple things at once, something I can't even do. Um, and so uh it's actually running these things in parallel while being able to answer new questions and even kick off new work. You'll notice that it actually messaged uh these other agents. So I can pop open this chat and see what's

[34:47] happening back here. Um they're chatting uh and I'll close this and then I'll show you this one here too. So also having a parallel conversation uh on accessibility here and so um it's actually going to check in on the bots for me and if it needs to redirect uh it'll do that and so it's really helpful to have just kind of like one entry point. Um I actually ran some of these already for you all so that we wouldn't

[35:13] have to sit here watching the paint dry. Um but uh let me show you what happened here. So accessibility Amy actually went in and found some issues. Um, so for example, um, here the, uh, it says down here, home hero image with no alt text outlined in red. So it put a little red border around that. So there's no alt text on that hero image. Uh, here it found um some news dates with low contrast. Again, this is looking for accessibility issues, and that is even hard to read for me. So [laughter] I I think like on a small screen or people who have you know issues with

[35:50] vision um they would have a problem reading that. And then also here the help center student ID field with no label um outline and red as well. So it found three accessibility issues and um it actually uh kicked off uh cloud agents to fix them. So uh I haven't even seen them yet and they're

[36:10] already fixed. So pretty incredible. also show you here um the uh so Grants Greta also came back with a uh report as well um and so you know this would have taken maybe days um to be honest I don't work in these departments but um with our reports internally sometimes they take weeks and uh it's able to do them um with extremely high accuracy sometimes in in minutes because it's hooked up to all of our systems and so So it's it can work

[36:42] really really efficiently. So okay um I want to show you one more thing here in accessibility Amy's uh chat. So um one of the big reasons Grockbot is able to do all these things and act as a real teammate like I said is the fact that they're always on. So that just means that they have their own computer and they have the necessary tools to get things done uh on their own. And so, um, actually what happened here, I'm gonna

[37:12] open up this computer. And what accessibility Amy did and maybe is doing right now, yeah, is actually doing right now is it's um checking the entire website for accessibility problems. And, uh, it's doing that in two ways. Uh, one is doing it visually the way that a student would actually experience the site. So it's actually going through clicking around and like

[37:36] with its own eyes identifying issues. The other cool thing it's doing is it's simultaneously checking the code on the front end. So it's actually writing code into the console here. It's checking the HTML. It's checking the CSS. And so it kind of has X-ray vision in that way. Obviously an engineer could do this too, but it's painful work. Uh it would take a very very long time. It's able to do this all on its own. and learn how to navigate the site more and more effectively. Um, while we're here, I just want to show you one kind of cool feature here that doesn't apply to this particular

[38:12] workflow, but I can actually take control. And so, uh, with very little latency, I can say like, you know, I can actually test things out on its computer or, uh, make changes on its computer at any time and then hand back the control and it knows how to pick back up from there. So, um, that's one of the major differentiators there, uh, with probably some of the other AI products that you've seen is that they can actually

[38:36] click around and do real work. So, uh, another cool thing about this is that you can actually teach it a task. So, for example, I can, uh, hit this plus button and I can say teach a task and actually start recording. And so if there's something that I want done a certain way, um, I can go in here and, you know, click around and explain essentially the way I would with a new

[38:59] hire. Hey, here's how you do this thing. And then I can, you know, stop the recording and it'll watch the recording and learn how I like something done. So if you have complex systems where first you have to go to this one system and copy paste this thing and then like plug it in over here things like that that might be really hard to explain. You can actually just teach it the task the way you would a human being. So, um, okay, the last thing I'll say here is that I can actually say run an audit of our whole website weekly and fix small copy changes yourself and

[39:38] surface bigger issues to our team for review. So again, this is extremely helpful, but it's I'm we've been treating it kind of like a one-off when in reality looking for um ADA compliance or any other kind of compliance is actually a routine task. It should be happening, you know, every so many days. And so here it's

[40:02] actually setting up a uh weekly audit. And uh you'll notice up here in the top right that it has this um upcoming uh run on Monday at 8 a.m. I can tell it to change it. Hey, run it on Fridays or whatever I want to do. Run it daily. And it'll edit the routine over time. And as you start to teach it more and more, it'll actually baked bake those

[40:23] learnings into that routine. All right. So, let's go over to um the help desk. So, let me pull this up here. All right. So, um, obviously students submit tickets all the time. I'm just going to fill one in, uh, really quick here. So, I'll say Santi Garza, and I'll select a category. Let's do financial aid hold and other

[40:49] hold. And I'll do, uh, bogus student ID. And that's my cursor email. Um, let's sayars.edu. and I'll say uh hold blocking registration and then I will copy paste this in. All right. Um and I'll say urgent. So I've submitted a request here. Um Oops. Guess the form validation works. Let's see. Nine numbers. Am I there? Cool. All right. So, I've submitted this ticket through the website and um so let me show you something pretty incredible. This ticket was just processed by the system on the front end, but you'll see that intake Andy is

[41:41] actually working over here on the left. Uh it's moving around and it's actually handling it uh on its own. So, um the way I did this, by the way, for anyone interested, is I just asked Grockbot to set up a web hook against this form. So for anyone who's technical in the room um or not technical, Crockpot can figure it out. Um I basically said, "Hey, listen to this form input and when a ticket comes in, just handle it. You know how to handle it. Obviously, I've trained it already." And um and so it was able to process the ticket on its own and it even messaged its manager. Um

[42:17] so you know, if something very serious gets propagated up, I'll I'll see it. Um so that's pretty awesome. Um, but uh let me show you one more version of this. So, I'll go back to let's go to the housing page this time and I'm going to call support and hopefully you guys can hear this. University of Mars housing. How can I

[42:41] help? Hi, I'm Santi Garza. Uh, and um I'm here on campus. The heater in uh my hall is room 214 has stopped working. Nice to meet you, Santi. Could you share your student ID, M email, which hole you're in, when you're free for a visit, and how urgent this is? Sure. My student ID is 700123456. My email is Santiar.edu and uh the heater is in Isidus Hall,

[43:18] room 214. It's urgent. Just your availability left. What times work for you?

[43:45] Okay, so you just saw a lot. Um, but uh it filled out the form in real time so that I can review uh and edit, you know, any misspellings that might have happened there. It also uh made sure to gather all of the necessary information uh the way that a human caller would. Uh, Lindsay, were you guys able to hear

[44:05] that on your end? You were. Okay, cool. Um, it filled in the categories, even the drop down categories, um, and, uh, gathered all the information for me. It also, uh, is able to handle any language, um, which can be very difficult for call centers to do because there are a lot of languages out there. And so some of these um you know for international students um it would be really helpful to be able to communicate in their own language and it's able to

[44:35] um handle that seamlessly. Um but uh but anyways let me go ahead and submit this request. Uh by the way you could have had it just submit the request on its own. I just set it up this way so that I could review it. Um but uh I just want to kind of really quickly mention that as part of the SpaceX SI platform, you do get access to things like this out of the box. So the Grock voice API is uh currently number one on speechtoech on artificial analysis and most other benchmarks at the time of recording. It is an unbelievably good voice model as you

[45:12] just heard. Um it it just handles e everything. I mean, I've tested it in demos and uh that's why I'm comfortable running a live demo on it. It's it's just that good. So, um I just want to say that here at Starlink, we receive about 20,000 uh support and sales calls per day and it's able to handle roughly 70% of them without any human intervention um using Grock voice. Actually, the reason that there's that other 30% is because we actually want to speak uh to customers in certain cases and it knows intelligently to surface that up to a human. Um so, uh you know, it's it's

[45:51] it's I think it would be great for any higher ed institution personally and you could hook it up to Grockbot. So, let me just show you real quick. Um it was actually able to uh here's a phone call in Spanish that came in. happened while uh I was on the call. It was actually able to kick this off to uh a crockpot agent and which then messaged the uh student services manager. So all this is getting handled and it starts to work kind of like a department kind of like a

[46:19] machine under underneath you. Okay, how much time do we have? Not that much. I'll be quick. Um let's talk about cursor. So um for anyone technical or not technical, cursor is unbelievable. unbelievably powerful. Um, it's used by 64% of the Fortune 500. It writes a 100 million lines of enterprise code per day uh at the time of writing. And um on campus uh it gets used a ton as well. A Stanford team with no webdev experience recently built um Alphacive with Cursor which is a scientific discussion platform and now hundreds of thousands of researchers uh use it today. So, um, really, really powerful if you want to

[47:06] build applications. Um, I'll just say I'm going to go over to Intake Indie and say, "Hand the website fix to a cursor agent. It's a texton change. Show me a screenshot of the fix page." Now, this is referencing something that happened earlier uh where there was an issue with financial aid where there was just a typo. It had the wrong year. And so, uh, it's going to check in and it's actually going to kick up a cursor cloud agent to handle the coding change. And, uh, like I said here, it's messaging, um, with this time not another Grockbot, but it's actually messaging with a coding agent. Um, if you're interested,

[47:48] by the way, in cursor and cloud agents, I highly recommend you go check the uh, workshops page. It's cursor.com/workshops. And let me just show you real quick. Um, this is a really great stop. Such a rich resource because not only do we run these live sessions, we actually um uh record and and publish them here. So, you're able to go check on any specific um or general category. So, for example, I highly recommend cursor 101, cursor

[48:19] 2011 for anyone interested in cursor. So going back here, um it's interfacing here with the cursor cloud agent and actually it's going to kick up the the agent and then check in on it. So it has access to its agent trace and it can follow up. Um like I said, I already ran this and it came back to me with a screenshot where it was able to change the the typo change. And I'll just show you um the the PR here

[48:45] if I can find it. Um, no, sorry, that's the wrong one. Well, anyways, um, basically if it's a small copy change, I've set it up to automerge and auto deploy. Um, so if it's a really small thing, um, you can set up that way. Uh, I'll show you. I know we're out of time. I'll just show you a few quick cloud agent runs just for fun. Um, let me pull them up

[49:13] here. So, let me pull into my other account here. One sec. All right. So, uh, last night I had it actually run, let me pop this open a little bit more, a few more ambitious tasks just so you all could see. Um, so here it created an entire site map on our website of the university. Um it also uh created a in this case it created a um like a command K like search everything kind of

[49:51] intelligent search for the application. So up here in the top right you can actually search pages and you can see that it's demoing it on its computer recording the change and actually pulling it back for me to review which is just incredible. Um I'll show you um a couple more here. There's some fun

[50:09] ones. Um, let me find Hold on. Um, where are you? This one's kind of cool. So, this is an interactive uh Mars globe that it built on the website. And um, if we had more time, I would show you more changes, but it it this is all cursor and it's able to do really incredible engineering work. Um, and like I said, Grockbot acts as kind of like an outer loop, kind of like a manager uh for engineering tasks. And so even if you have no coding experience, you can actually um have Grockbot handle

[50:46] and manage this stuff and build it out. So um the other cool thing about it is if you do have engineering experience, uh you can actually build systems uh that review uh for bugs, review for security vulnerabilities. You can run like mythos style um security audits. You can um we have a new feature called rollouts which follows your code into production and checks uh against your uh sentry and and you know basically all of your logging to see if it had a positive and or negative effect on production. So

[51:21] there's so much you can do with cursor. Um check out Lauren Tan aka Potato. Uh she has some great videos on how she has set up um cursor to run autonomously. All right, I know we're close to time. I want to leave time for questions. Uh so I'll stop there and um yeah, Lindsay, do you see I do. I wrote down a couple. I'm going to start actually with the compliance piece like out of the gate. I think it's really important to understand how we're compliant in this space. So SpaceXs SI we are almost at our Fed ramp moderate moving towards uh also a Fed ramp high

[51:55] compliant if that's important to you which some researchers and R1 with DoD funded data that's important to them. We also have a BAA for everything on our platform except but we are working on that. It should be here very shortly. I don't have a firm ETA but we're almost there. It's just a newer product. So, of course, getting these compliance and audits take a little bit of time as we're kind of moving as fast as humanly possible. So, there was a question about HIPPA FBA and Grockbots's the only piece of that puzzle. What I've seen universities do is still test uh test the platform and come up with these use cases, but we

[52:33] just make sure we don't include HIPPA data within the Grockbot piece. Another uh question that was asked was um like just the platform overall and different use cases and just how we would test it. Typically we have universities that do or come up with a couple use cases and you can test any piece of our platform. So Santi had talked about the four pieces of SpaceX SI. You would be able to test each of those. So cursor or grock build, grock API, Grock main that truth truth seeeking chat and Grock bot all within the platform. Four pillars that we help you walk through specific

[53:09] use cases hold your hand the entire way. The last thing I'll say then I'm going to pass it to Santi. I have a couple technical questions for you. Um people are asking about token optimization and um token base. So this is uh we we typically come up with token spend or a we agree to a bucket of tokens that you can use any piece of our platform. So it's not a per user per se license but you would essentially commit to XYZ bucket of tokens and you can use any piece of our platform and just submit it against the tokens. All right that was really quick. Um, Santi, can you go into

[53:47] a little bit? Well, first off, can we submit bots to a public marketplace? Yes, you can actually. So, you can submit bots to a public marketplace. Um, and uh I'll have to pull the exact if you look it up, you'll find it. I'll have to pull the exact uh form, but the the marketplace that I showed you there uh briefly is actually um let me pull it

[54:09] up again. There it is. So um you can think of this as like kind of like our app store. These are reviewed by us when you submit them and certain um bots are published. Something I didn't mention because it is on some of the other workshops is that you can actually uh publish a bot um to your team. So if you have a team account, you'll actually be able to distribute it internally. And this is really really helpful. Uh that way everybody doesn't have to create their own version of a bot if you have something that should be

[54:43] department or institutionwide. Awesome. Um how long do you think it's taken you like in general to spend time with your bots and invest in your bots to help with these like complex tasks and workflows that you were showing and demoing? Yeah, good question. So, uh, I'll I'll speak about my personal my my real work, uh, life as opposed to this demo because the demos are kind of, you know, they're a bit thin obviously. Uh, they're just demos, but so, um, Grockbot is remarkably helpful out of out of the gate, like day one. It's very very good at um, figuring out what you need and inferring that and like getting

[55:31] right to work. uh we uh we are first movers in the space and so we're actually able to iterate quite a bit um uh before launch and it's it's they've gotten remarkably good. Now just because they're good doesn't mean they're doing things exactly as you want them to to do them. Uh they can't read your mind and so there's quite a lot of preference. Um and so uh oh man, how long has it taken me? It just keeps getting better and better by the day um because they they have that memory and adaptability. So,

[56:03] I mean, it's a hard question to answer. Like I said, every day literally gets better. Um, at this point, they're essentially running my life. I'll admit, uh, we we internally, I'll just say real quick, internally, we we do strongly encourage you to take like a review approach to their work early on, especially until you build out those

[56:26] systems. Great. What about bots as code? So, essentially manage and disseminated. Yeah. So, we actually have a I won't have time to get into it because we're almost at time, but we do have an SDK, an agent SDK. Um that is essentially bots as code. Um it's really a structure for you to um be able to have evals and hill climbing. And so you're if you want something more structured, um I highly recommend you check out our agent SDK

[57:02] for that. Awesome. And last piece I'll add, I know we're right at time here. You guys are asking a lot about the workflows. Don't get so hung up on the piece that Santi was talking about that's not public. I have a one-pager that is public. I will maybe I can put it on our website and it has just different use cases that my higher eds are testing. I cover all higher education within the US. anywhere from large IVs and R1s to the really small universities. So, we have we had Grockbot scrape all of my calls um for all of the use cases and you'll be able

[57:34] to see the use cases and then you just pop that into Grockbot and ask it to work it out for you and we'll show you a flowchart. And the best way to do this, you can um get in contact with me. I'll maybe we can follow up with contact information and the best way to um see is to use and trial our products. So, we're always open to trials with specific use cases. So, thanks for hopping on. I think that's it. We're right at time. Santi, great timing. Down to the second. That

[58:01] was perfect. Thank you, Lindsay. Thank you everybody for joining from all over the world. I hope this was helpful. Please do check out our cursor.comworkshops page for more and sign up for uh for more upcoming sessions. And thanks for all the reactions. I do see those and they do keep me going, so I appreciate

[58:18] it. All right, all um have a great day. Have a great night. See you all later. Thanks. Thanks you guys. Bye.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt