Canvas Next/IgniteAI Agent Demo Video
Instructure's team demonstrates the Ignite AI agent, a key component of their Canvas Next tier that integrates AI tools directly into the LMS.
Welcoming the team from Instructure. Over to you all. Welcome, everybody. Excited to have you all here today. We're gonna be showing off one of the things that I'm most excited about we've rolled out recently, and that's our Ignite AI agent. It's one of the many pieces of our new Canvas Next tier.
Quick round of introductions. My name is Daniel Guillo. I'm a solutions engineer here at Instructure. I'm joined by Zach Pendleton and Olivia Quito, which I'm gonna give them just a moment to introduce themselves as well. Hi, all.
My name's Zach Pendleton. I'm the chief architect here at Instructure. Thanks so much for joining us this afternoon. Hey, everyone. My name is Olivia Quito, and I am a product manager at Instructure working on the agent.
So super excited to show you some of what we've been working on today. Thank you all. And you all are gonna learn a little more about me later, as we get into some of the demo. But real quick, high level, what we're gonna be talking about today, Zach's gonna kick us off with some common pain points and challenges as institutions are beginning to roll out kind of faculty level AI solutions. We're gonna talk about the advantages of putting your LMS at the center of that AI deployment.
I'm gonna go into a demo of specifically what the agent can do, and then Olivia is gonna come in and talk to us about what's next, and we'll do some q and a. So take it away, Zach. Alright. Thanks so much, Daniel. You know, we've had a lot of conversations with institutions, that are trying to make sense of what AI looks like for their students, for their faculty.
And while every institution's desires and needs are a little different, we think that what's true across institutions is that they're trying to strike the balance here between feeling the need to do something, but making sure that they're doing it responsibly, and in a way that serves education, and isn't just deploying technology for the sake of deploying technology. So and and that that comes across, as we look deeper into what particular challenges folks are having with AI deployment. Right? Certainly, there are some technical challenges, but most of the challenges are are about trust. They're about faculty buy in, and they're about ensuring that these new tools can be used in a way that, is fair and ethical. And so to match those needs, as much as we're seeing schools ask for technology solutions, they're also asking for, help in understanding what a good AI tool looks like.
They wanna have discussions about what, human oversight and human in the loop looks like as well. And so as we've built the entire Ignite AI platform, we've focused on answering some of these questions, for folks and and making sure that that it exists in a way that is transparent, that is clear, and that is focused on the classroom. And so for the first question, right, schools often ask us, how do we know that an AI tool is appropriate for our use case? Similarly, they ask, how does our staff know that's available once it exists? Which I think gets to Daniel's point about having the LMS right at the center of things. And and then making sure that these tools are focused again on the classroom, on real and and present needs and are not just technology for technology's sake. And so Ignite AI tries to, tries to help with all of these problems.
You know, first with the question of vetting, we, you know, we provide, robust documentation, about what our features are, how they're used, what the expected risks, outcomes, and are. I I think is there a is there a click through here, Daniel? This should be Or is that it? Answers. Yeah. That's it. Okay.
Perfect. Good. Alright. Adoption, having Ignite AI right in the LMS means that you can see very clearly what tools are being used, and people can see, those tools popped up right in their existing workflows. And because they're right in the existing workflows, right, we know that they're delivering, value and and helping to simplify the experience folks are having.
So, with that, I'm gonna hand it back over to to Daniel to to do some demoing. Thank you, Zach. And I'm gonna set a little bit of the stage beyond that. So, again, we're gonna be talking about Canvas next. And, specifically, that's our tier that's very AI forward, essentially.
All of our ignite AI tools are going to be available in that tier. The capstone is definitely the agent, but there are a lot of other Ignite AI features. I'm gonna mention some of those in passing, but they're not gonna be the focus of the demo today. So just kind of setting the stage there. But let's talk a little bit about having your LMS as kind of that center of your AI deployment and what those advantages are.
So Zach's already kind of talked a little bit about this. You have a lot of transparency, and I'm gonna show you once we get into the demo. Pretty much all of our Ignite AI tools have our AI nutrition facts attached to them where you can see what model's in use, you can see how the data is being used, and more. There's configurability. You've got control over what tools you make available to your faculty, to your to your students, so you have that ability to keep things where your comfort level is, where your staff's comfort level is.
Your tools are in context. Right? So our agent, in particular, we're gonna see is aware of what I'm looking at when I do the demo later, and I'm not having to move content between different platforms. I'm not having to go copy my lesson into ChatGPT to get feedback, to copy back over to Canvas. It's all in one place. Right? That agent is going to be that conversational assistance.
We're used to this with LLMs. Nothing crazy new there. But I don't have to worry about technical terms other than maybe a few just key specific Canvas terms just to make sure that things are right, but, again, very conversational. And then educational value, it's really streamlining that workflow, keeping everything in one place, giving you more time to focus on students and help ensure their success. And finally, right, we have many robust integrations in Canvas.
As our agent improves, they're certainly going to be taking advantage of those connections. So that's a lot of talking, and I've got a poll really quick just to get a feel for kind of how y'all are most excited about using AI tools at the moment and what would be most valuable for you. But I promised a little more about me at some point. So I've been at at Instructure for three months now or just about. This week rounds out my third month.
So I'm one of the relative newbies here. But prior to joining Instructure, I spent some time in higher ed, some time in k twelve, some time in instructional technology. And that in that last role, I actually had the opportunity to be a part of our early adopter program for the Ignite AI agent. So I was one of the first people to get my hands on the agent and fell in love immediately, which kind of surprised me. Historically, I've been a little bit of an AI skeptic.
For a while, my stance was kind of really the back and forth here. I'm gonna spend more time getting the the LLM to produce what I want than if I would have just built it myself. But the agent kinda changed my view on that, and the big big piece of that is because it has access to over five hundred API endpoints inside of Canvas. And that's not just pulling data, that's actually executing things and and taking action on my behalf within Canvas. So that's going to include things like changing due dates, creating assignments, creating pages, populating those things with content, sending messages to students, managing differentiation tags, and more.
And that list is growing steadily. Our team has been amazing since I've been a part of this since very early. It's been fascinating watching how that has grown and how it's improved. Things that used to take minutes now take seconds. So that's a lot of talk.
Let's actually get into a demo, and we can go ahead and close that poll as well. I'm curious to see where what y'all are interested in. Alright. So content creation seems to be one of the big ones. We've got some data queries and insights.
Alright. Good news. I've got examples lined up to cover the, the big things here. Now I did mention, there's gonna be a focus on the agent. We do have I do wanna highlight a couple of the other Ignite AI tools.
We have a rubric generator. We have, item authoring assistance for quizzes. We have grading assistance. Right? We have several other dedicated tools. The agent is kind of our bigger, able to handle lots of complex task tools, which we'll eventually be passing off to some of those more specialized tools.
So I'm gonna jump into Canvas and show you a few things here. First and foremost, how easy it is to get to the agent. This lovely button is going to exist at the top of just about every page in Canvas, makes getting there to what I need very easy. I wanna point out one of the very new things our agent team has released is our community library. This is a nice public space, for our users to share prompts that they found are very impactful for themselves with others or to go get inspiration.
So huge kudos for our team for rolling that out recently. I'm gonna get into the AI nutrition facts in a little bit just because I'm gonna need some filler as things are processing, but I'm definitely gonna show you that. And I'm gonna dive in and show you one of my favorite content creation pieces. I'm going to add a disclaimer. This page that I'm about to pull up is going to maybe be a little overwhelming.
It's a wall of text. And my second disclaimer is I teach calculus, and I know people have strong opinions about math. So ignore the content. Focus on kind of what the agent's doing. But this is something I've been doing in my course.
Right? Because I'm still teaching. I've I refuse to let them take that class from me. I enjoy teaching calculus. So here, right, I've provided a transcript of recordings that I've been doing. Right? So I've been recording a lot of content from my course, and it would be amazing to have a text version of that to go along with my video.
Right? Because I know me, I'm old school, give me text anytime over watching a video. I know that's kind of some generational things there. I'm probably showing my age. But it's very easy to come in and ask the agent to give me a lesson. Right? So I'm asking it simply for, hey.
Here's a transcript. Give me a asynchronous learning activity that's aligned to this content. It's gonna take it a minute because it's gotta process what's here, and you can see that's kind of an ugly file. So I'm gonna take this opportunity to highlight those AI nutrition facts. Right? So you can see here the agent currently is based off of Haiku four point five.
If you wanted to see data permissions and those sorts of things, all of that's readily available here as well. Also, I'm gonna add a disclaimer. This is, you know, an AI agent. It may make mistakes. It's it's been acting pretty well.
But every now and then, it decides to go a little crazy. I don't think we're gonna have any issues today. But just to kinda set some expectations, right, I like to think of it as a very enthusiastic TA. Right? It's a great assistant. It does need some oversight.
So you will have to keep an eye on it. You always wanna verify what it's doing. You just saw that lovely wall of text, so it's done. Right? So it has built for me a nice lesson, and the agent is always going to keep me informed. It's going to show me what it's getting ready to do.
It's getting ready to make a page with this content, but it's always human in the driver's seat. And it's always going to ask for my approval before it actually completes any action. So I'm gonna approve this, and it's already got everything done, so it should take it just a moment to make me a page. But while I'm waiting on that, I also have access here to my entire history of chats. So if I needed to come back and resume a chat, continue, iterate on something, I do have that opportunity.
So I've got a direct link to that lovely new thing that it made, so let's go look at it. And from that ugly transcript, it has built a decent lesson, and I can verify that these are exactly the examples that I went over. You know, I probably wanna go clean up the math notation. That's a personal problem, but I would call this probably about a ninety percent good lesson. Now next thing you might be thinking to yourself is, well, that's great, but my institution has a template that my content's supposed to be formatted like.
Or I have a personal look, and I have a personal style that I want this to look like. So I have another lesson that I've built that this is really what I want everything to look like, keep my class nice and consistent, all of my coursework the same. So I can very easily come back to my agent and ask it to create a copy of this and really format it more like that sample. And I'm gonna give it some time to run-in the background. I'm actually not gonna let this one complete because it's gonna take a while, and I've got more things I wanna show y'all.
So I ran this previously, and this is that result. Right? You can see it did a fantastic job of taking that lesson that it built and really making it follow that that blueprint of my own. Now as far as, like, blueprints that you might have at the institution level, right now, this would be something that you would need to have your instructors import in their course, but then it would very easily be very easy to point the agent to that. And this has been I I have not manually edited anything. Right? All of this has been strictly prompting the agent, essentially, take this, produce this.
And you've seen the prompts that I've used to do this. So it's very intuitive. It again, it might occasionally decide to do something a little weird. It's pretty easy to reprompt. Hey.
This doesn't look quite right. Can you try that again? This part's messed up. It's also very easy to conversationally, hey. You added a sixth section here that I really don't want. Can you just remove that piece and make those quick tweaks without having to really dive into the weeds? And this is all well and good.
That's a lesson. Wouldn't it be great if I could do a lot more, though? So, again, staying aware of what you probably already have in Canvas. Right? So, again, context aware, every single one of these prompts, I've said something like this assignment, this page. The agent knows what I'm looking at, where I'm at. Right? So I've got a course outline.
Right? This is one of those things that more than likely you already have in your course anyway. Right? You can see my lovely, slightly complex prompt here asking the agent to take chapter three, which is eleven sections worth of material, and go make a module for my chapter, an assignment for each lesson. I gave it some loose parameters for due dates. I could have added more. I could have told it specifically how many points I wanted things worth, given it more parameters for what to build, but I kept it simple and let it run.
Now this is another one of those that I ran in advance because I wanna show you exactly how thorough it is, and it's going to show me what it's doing in every step of the way. It's going to go create my module. So, again, taking action for me, And then it's going to start building my assignments. And it's going to stop and ask for approval at each step, and then it's gonna go individually move each of those assignments into the module I wanted. Right? So there's a lot of babysitting right now to approve, approve, approve, approve if it's a multistep process.
I will say one of the things that our team is working on to steal a little bit of Olivia's thunder is being able to batch approve this so that I can just approve once for kind of a bulk action once it's given me that game plan to help reduce a little bit of that friction. And using that, it built out my module. And diving into one of those lovely sections, again, is aligned exactly to that template that I wanted. So content creation, the agent is amazing for this. And and I know one of the other pieces to seal again a little bit of oblivious under is going to be eventually the ability to upload and attach files to that that, prompt that you're doing.
So if you have a PDF of your syllabus sitting somewhere, being able to do that instead of necessarily having to have it on a page in Canvas. So that is something that's coming. So that's a lot of content creation. Oh, and I'm trying to keep an eye on the chat. I see a few questions that are gonna be good for later, but I see the one about CityLabs that just came in.
As a former CityLabs user at the institution I was at, right, that was the kind of template I was having to follow, and it can handle that very well. So I can confirm that one. That was actually the use case I was most interested in originally. So that's a lot of design content creation. One of the other things that I think is most impactful here is going to be really helping do some differentiation and making things a little more personal for students.
And and, also, I wanna touch a little bit just on, like, the data and analytics. So here, I have you notice this one's not a fancy one. I haven't cleaned this one up with my template yet. But I have an assignment that some students have done, and I wanna know which students scored under a seventy percent on this assignment. And you might be thinking to yourself, well, I could very easily go to my Gradebook and see that.
Right? It's very easy to go sort by grade, high to low, low to high. There they are. Move on with my life. At least if you've got a small enough class. You got a big class that's gonna take you a little longer, but I've got my students.
But, really, the magic now is what I can do now that I've identified these students. I could create a differentiation tag and put those students in it. Right? Maybe I wanna track them that way. I could do a little more analysis on their current grades in the course. Or what I want to do in this particular case.
And the assignment result here is not going to be amazing because my prompt is very basic, mostly because I wanna keep this quick. But I could ask it to create a remedial version of this assignment. Right? Something that's gonna point out common mistakes, misconceptions, add some additional scaffolding. Right? But I just wanna assign it to those students that I can tell need a little more assistance. Right? I want to be able to ensure their success.
And so it's made that assignment. I gave it some specifications on the amount of points, and it's adding those overrides. So I don't have to go find those students in the list to manually add them to that that exception. Right? The agent's taking care of all of that for me. And now if I also told it to publish and did some more, I could take this a step further.
And let me grab my prompt. And I wanna send a short message to these students, letting them know they've got an opportunity to improve their grades. And, again, the agent is taking action on my behalf. So when it sends this message and I can always come back and review it. Right? Okay.
That looks like it's probably not too bad. Go ahead and send. Or if I don't like it, have some back and forth. Might just say, here's the message I wanna send if I know this is really what I want to send. Right? That message is going to get sent individually to each of those students from me.
Right? The agent is taking action on my behalf, which means those students are going to see it come from me and can reply to me if they have additional questions. So, again, very easy to stay involved, to stay connected, and have that human in the driver's seat. Now from here right. So that's kind of that really bringing you more time to do what matters, and that's have that student impact. This is another of those use cases that I'm most excited about is that whole let's build some differentiated lessons.
I need a version of this that provides additional support. I need a challenge version of this. I've got some high achievers that I wanna kinda push a little further. It really kinda depends on that workflow, but there's a lot of options to really generate a variety of content quickly. Again, TA, always go double check behind it.
Don't just let it run off on its own, but it's there as a great partner. Now the the last thing that I wanna look at is the feedback piece. Right? And this is really good, not just kind of at the instructor level, but this is going to start being something that's valuable kind of administrative level. Right? I imagine a lot of, institutions do some version of, course reviews where you're gonna go review content in a course looking for certain things. The agent is great at that as well.
I'm gonna give kind of a generic set of directions here and ask the agent to just kind of act as a instructional coach for online instruction and give me some some feedback on my assignment, giving me what's working, what's not working, maybe what I can improve. Right? But, obviously, if you had some kind of institution level rubric that you wanted to kinda feed in here, if you followed something like QM or the NIT rubric, you could just reference those by name. And it's gonna give you feedback then on what's on that page based off of what you've asked for. And this should be pretty quick, and this one's generally pretty reliable as well. And there we go.
And so it's given me what's going well. It likes my learning outcomes, my scaffolding. It's pointed out I don't have a rubric. Yeah. I should probably do something about that.
It's pointed out a few kind of Canvas side specific things. Maybe it's not published. Maybe I'm hiding grades. Is that are those settings intentional? Maybe the due date looks a little weird. So it's not just looking at the content on the page.
It's looking at this item in Canvas holistically. Right? And so this is, again, one of those huge advantages of having such a deeply integrated AI in the LMS, right, is it's really not just surface level. It can go a whole lot deeper, and I'm not having to spend time jumping between platforms, copying and pasting, files. Again, really saving time to do what matters the most and improve student impact. So that's really high level, what the agent can do, some of my favorite prompts that I've been using in my course.
And with that, I'm gonna pass it over to Olivia, who's gonna give you all a nice little preview of some of the things that they're working on. Alright. I'm super excited to show you on top of what you've seen from Daniel that's already working in the agent today. One of the many new things that we're working on for the agent that I'm particularly excited about. So let me just share my screen.
Alright. And what I'm gonna show to you today is called prompt builder. As you saw from Daniel's demo, the agent has infinite possibilities. You can ask it to do so many different things, and that's powerful. That makes it very open ended and powerful for many different use cases.
But we also hear from educators that that can be a bit intimidating. It's an empty chat box, and you don't know what you don't know, you don't always know what you want the agent to do or what it can do. So, we've been exploring, in addition to what's already there, having a more scaffolded experience that can help educators with a range of comfort with AI actually get value out of the agent. So, that's what you're seeing here. And this is a a live prototype that you're seeing.
So, this looks very similar to what will actually be in Canvas. And in prompt builder, you can see there are some different actions that I can take. These are all things that the agent can do, and it's gonna walk me through how to build a prompt to do these things. And, right now, there's a list of six things, but this is something that we anticipate we'll be expanding to be adding many more different use cases. So I'll just show you a couple examples.
We'll do scheduling and extensions. This one is quite a powerful use case for the agent. If for teachers who have a class where they need to do, like, bulk date updates or bulk updates of many different assignments or modules. And this this prompt builder flow is just gonna walk me through that. So I can tell it what course I wanna make updates in, and I get some some guidance here on, you know, how I might wanna refer to my course.
If I'm already in a course, it will autofill this. So it'll know what course I'm in and it'll put that in this text box. And then, this is where some of the magic comes in. I I have a a text box here where I can I can type in whatever I want? Do I wanna ship the assignments forward seven days? There there are some examples that I can I can go through here to sort of get my imagination flowing? Do I wanna push all due dates back one week? And, again, I have some some guidance here. So, this is telling me things that I might wanna include in my prompt to make it work better.
So I can say can edit any of these. So we'll just say shift all assignments forward seven days. It'll say all assignments in the class. So, this is not only walking me through it, but it's it's sort of teaching me how to build a prompt and how to tell the agent what I need. And then, I can press this generate Prompt button.
And this gives me my agent ready prompt that's going to do what I need it to do. And when this is live in Canvas, I'll be able to click this Insert Prompt button. And that prompt will go straight to the agent and it'll do the thing. So, it's right there. Or I can copy and paste the prompt if I wanna save it for later.
And I'll just show one more. So, we also have a tags and groups prompt builder flow. This one's a little bit more guided and that just depends on the task. So, for this one, I can choose what kind of do I wanna create differentiation tags or groups? What do I wanna do? Do I wanna create or manage them? And then, again, I have this text box where I can add all those extra details and specific things to my situation that I want to include. And, once again, it will give me my prompt.
And that is Prompt Builder. We're really excited to release this in the near future. And, I will be on the line for Q and A. So, happy to answer any questions that you have about it. Alright.
So we're gonna kick over into q and a. And we've got a few lined up. So I think this one's gonna be for Zach. Is there a way to set up Ignite AI to be a sort of coach or tutor for students in understanding a specific concept and clearing misconceptions, or will the Ignite AI agent only be for instructors? Yeah. That's a great question.
I've seen this a few similar questions here about student functionality. So we we do have a lot of other study tools available for students. The Ignite agent that we showed today is just available for educators, faculty, administration, and staff. Thanks. For Olivia, will the Canvas AI agents be standard across all schools and universities, or will each campus customize which AI tools it has agreements with and only use those? So, the short answer to that question is Ignite AI will provide a consistent foundation, but institutions will have control over what's what gets enabled.
And we are exploring building out MCP integrations within our partner ecosystem. And, as those are built out, you would be able to use the agent and a third party tool together. For example, let's say you use another EdTech product to build out interactive presentations. Once that MCP integration is built out, then you would actually be able to prompt the agent to do that, to to make those presentations for you through that external tool. Awesome.
And another one for you, Olivia. How many tokens would an assignment generation typically take? That's a great question. So, the agent is actually designed to support the vast majority of expected educator workflows without requiring instructors to think in terms of tokens or manage usage at that level. For many common tasks like generating assignments and updating course content, drafting communications, all things that you saw Daniel do with the agent, users shouldn't need to worry about running it to limits. That said, some AI workflows require significantly more compute than others, especially if they're involving very large courses or extensive content or repeated high volume actions.
And, we will have safeguards in place to protect performance cost and responsible use. We're also exploring the right ways to give institutions visibility and control over usage without making the experience feel overly technical for educators. So, this this will evolve over time. And, the goal is to make Ignite AI agent feel simple and useful for day to day work, but still give institutions the oversight that they need. And, Zach, I've got a couple more for you.
There's been a couple of questions around data training and protection of intellectual prop property and privacy. Can you speak a little bit to those concerns? Yeah. Definitely. So the Ignite agent and all other Ignite AI features inside of Canvas are not trained on, institution, or student data. And in in fact, that is, I I think for us, a real feature and not a bug.
It's, you know, we're happy to be able to provide things, that don't rely on that or or don't change the game when we think about student privacy or, protection of of IP. Secondly, I'll I'll say Ignite Agent and, and all of our other Ignite AI features are hosted, using large language models available, through Amazon Web Services, through their Bedrock platform, which means that all of the data that gets sent at prompt time to the model stays inside this the customer's same cloud. Alright. Thank you, Zach. And one more, I think I'm gonna kick your way.
Can each instructor install the Ignite AI agent? Yeah. So, that's another great question. The Ignite agent can be configured on, off, available, unavailable at the account level. Right? So at that kind of top level inside of Canvas. But then it's configurable all the way down through subaccounts and then into courses as well.
Because we know, you know, most folks have got some faculty that are really excited about AI and some faculty that are maybe a little less excited or are not quite far as far along on their journey. So you're able to provide them with with unique experiences. Awesome. Alright. Back to Olivia for a couple more.
Can you speak to, the ability for the agent to be able to take action across multiple courses? Absolutely. Yeah. The agent has access to all of the courses that you have access to. And even when you're not in a particular course, it can see those courses and look at those courses. So, there there are many cross course workflows that we've seen people doing.
For example, if somebody's creating a new course, they might say, this some of this old content from a different course. Pull it into my new course that I'm building. Or searches across multiple courses is another thing that we've seen. But but, yeah, certainly, the agent is able to do that. And then for Olivia as well, I've got a couple of questions around accessibility.
So one one was curious about the accessibility in the prompt builder. And then, also, just some questions around, is the content that the agent is going to produce accessible? Yeah. Great great questions. We know accessibility is top of mind for for many educators. And yes, the agent does default to creating instructional content with accessibility standards in place.
As Daniel has been saying, it's like a TA, you always wanna double check it. But it does typically generate accessible content. And to build on top of that, a common use case we're seeing is actually using the agent to make content more accessible. The agent can look at content in your course and it can scan it, see if there are any accessibility issues, and let you know about those things. And and what you saw in prompt builder is sort of a guided workflow for that.
And and I actually wanna tag in a little bit on that one. I had kind of a surprising use case, essentially asking the agent to do some review for accessibility. And it was a lesson where the instructor had brought in some stuff that was originally done in markup, and the agent actually flagged that there it looked like there were some relics of the markup that hadn't been fixed, which is something that most of kind of our normal accessibility checkers wouldn't have caught. So it's an actual nice extra layer on things sometimes. And for Zach, is there a way for students to know whether certain course components or content is generated with AI? Good question.
So, we we hear a lot of different opinions, from institutions, on this question. So some institutions that, you know, want really clear tags on on everything that's been generated with AI and and others that want to leave it up to the instructor. So, we don't have, the agent kind of tag or stamp its own outputs right now. What we've seen institutions do is is add some standard footer or instruct the agent when it's generating content to make a note about how that content was AI authored. And, Olivia, does the Ignite AI have any sort of institutional guidance or training that'd be taken into account in all of its responses and actions such as, like, brand guides and policies? Great.
Great question. This is a common request request we've heard from from several different institutions that we've been talking to while while developing the agent. And definitely let us know any kind of customization you'd want us you'd wanna see. This is something that we're actively thinking about and exploring and and certainly plan to have some some support for this in the future. And I wanna give a shout out to Olivia's team, by the way.
They're currently hosting regular sessions, I think, right now once a month to get feedback from people who've been getting in and playing. And it's been amazing to see the rapid improvement and and those suggestions that have come into play very quickly. So just I wanna shout out that team for the amazing work they've been doing. And I see one that I'm gonna take for the is this an additional cost or is it available automatically? So right now, the Ignite AI agent is free for US Canvas customers through June thirtieth, and for customers outside of the US through September thirtieth. After that, it's going to be a part of the Canvas next tier, so that'll be all inclusive at that stage.
And for Olivia, how are chats with the agent stored and and how long are they preserved for? Yep. I can take that one. So as as you may have seen in Daniel's demo, we do have that chat history button up on the top of the agent. And, if you click on that, you're able to see all of your chat history, which would be all of the previous conversations that you've had, all of the messages. It looks just like it did in the conversation.
That all is preserved. And, we're having some discussions on data retention. So, we're gonna align with different laws and regulations about how long those are stored. I think, at the moment, it's two years. So, you certainly won't see your chats disappearing for a long time.
And another one for you, Olivia. You mentioned external tools. Will this be able to integrate with textbooks if they're used as an external tool? Oh, that is that is a great question. I was just having a conversation with the team about this earlier today, in fact. We're currently building out our first MCP integrations with the agent with select partners.
And, it's definitely possible that in the future, we'll have integrations built out with text providers. I think that would be very exciting, but not available currently. And for Zach, will the Ignite AI agent be able to identify accessibility issues and fix them? I understand the current Canvas course wide accessibility checker is limited to the Canvas HTML pages. Doesn't check or fix PDFs, will the agent do more with accessibility? Yeah. So I I think my answer is a nondeterministic yes.
Right? I think the the agent knows quite a bit about accessibility. I think it's able to make a lot of fixes, be able to coach and assist on those. You know, but because it's, it's not an on the rails experience like the Canvas accessibility checker, you know, it it may miss things, or it it may overstate some things. But I I think, generally, it does a pretty good job. I I would also say as regards to Canvas accessibility checker, that's another one to keep your eye on because we're, we're always making improvements there as well.
And for Olivia, this one's kind of a follow-up from a previous one, I think. Is the context of the agent limited to a single course at a time, or, is it context for everything you've asked across all of your courses? Great question. So the agent knows what course you're in. So, you know, for example, if I'm on the homepage of my calculus course, it's going to be primarily looking at that calculus course, unless I tell it to look somewhere else. But it does have access to all my courses, and it does like, I don't need to be on the page for it to look at those courses.
So, you know, if I ask, go back to my spring twenty twenty four calculus course or, you know, look across all my courses, it will be able to do that. I think there was also part of the question is, does it have context about everything you've asked? Right now, each conversation is separate. So there's no memory of you know, if you start a new conversation, the agent isn't going to remember what you asked in previous conversations. Although, that is something we've been having discussions around, and we may be adding that kind of memory in the future. And I see one that I'm going to take.
There's a question about sending messages to all students who have under a certain score on a specific assignment. Absolutely. That's one of those workflows that I I showed, like, the very last one I did or one of the last ones I did. You can certainly send messages to assignment based triggers, show me the students who scored under a certain threshold, over a certain threshold. You could also do that as a core score.
Right? I can look at holistic core scores. If I wanted to do a little more diving into some other data points, those are available. Essentially, as long as the agent can identify the set of students, you can target message those students. And I see one more that I'm also gonna tackle, and I think this is gonna be our last question just because we're coming up on time. So I'm gonna go ahead and get our little QR here up as well.
But I just little bit of clarity on, the the agent. I did say earlier that it's free until, June, and there was some questions about deployment. So it requires an admin to enable it. It's a feature flag right now for an admin to enable. So you would need somebody at your institution level if you've got Canvas to go enable that flag.
But, otherwise, that's essentially all that you need to do that for for that free trial period. After the free trial, then that's gonna require a subscription to Canvas Next. And if you want some more information about the tiers, that's what this QR code is gonna get you to. It's some additional information on the tiers of Canvas. And if you're interested more, certainly reach out to, to your instructor rep, and they can get you some more information as well.
So thank you everybody for your time. I think we're pretty much coming on time. It has been an absolute pleasure. Hopefully, that's got you excited for some of the things that we're working on.
Quick round of introductions. My name is Daniel Guillo. I'm a solutions engineer here at Instructure. I'm joined by Zach Pendleton and Olivia Quito, which I'm gonna give them just a moment to introduce themselves as well. Hi, all.
My name's Zach Pendleton. I'm the chief architect here at Instructure. Thanks so much for joining us this afternoon. Hey, everyone. My name is Olivia Quito, and I am a product manager at Instructure working on the agent.
So super excited to show you some of what we've been working on today. Thank you all. And you all are gonna learn a little more about me later, as we get into some of the demo. But real quick, high level, what we're gonna be talking about today, Zach's gonna kick us off with some common pain points and challenges as institutions are beginning to roll out kind of faculty level AI solutions. We're gonna talk about the advantages of putting your LMS at the center of that AI deployment.
I'm gonna go into a demo of specifically what the agent can do, and then Olivia is gonna come in and talk to us about what's next, and we'll do some q and a. So take it away, Zach. Alright. Thanks so much, Daniel. You know, we've had a lot of conversations with institutions, that are trying to make sense of what AI looks like for their students, for their faculty.
And while every institution's desires and needs are a little different, we think that what's true across institutions is that they're trying to strike the balance here between feeling the need to do something, but making sure that they're doing it responsibly, and in a way that serves education, and isn't just deploying technology for the sake of deploying technology. So and and that that comes across, as we look deeper into what particular challenges folks are having with AI deployment. Right? Certainly, there are some technical challenges, but most of the challenges are are about trust. They're about faculty buy in, and they're about ensuring that these new tools can be used in a way that, is fair and ethical. And so to match those needs, as much as we're seeing schools ask for technology solutions, they're also asking for, help in understanding what a good AI tool looks like.
They wanna have discussions about what, human oversight and human in the loop looks like as well. And so as we've built the entire Ignite AI platform, we've focused on answering some of these questions, for folks and and making sure that that it exists in a way that is transparent, that is clear, and that is focused on the classroom. And so for the first question, right, schools often ask us, how do we know that an AI tool is appropriate for our use case? Similarly, they ask, how does our staff know that's available once it exists? Which I think gets to Daniel's point about having the LMS right at the center of things. And and then making sure that these tools are focused again on the classroom, on real and and present needs and are not just technology for technology's sake. And so Ignite AI tries to, tries to help with all of these problems.
You know, first with the question of vetting, we, you know, we provide, robust documentation, about what our features are, how they're used, what the expected risks, outcomes, and are. I I think is there a is there a click through here, Daniel? This should be Or is that it? Answers. Yeah. That's it. Okay.
Perfect. Good. Alright. Adoption, having Ignite AI right in the LMS means that you can see very clearly what tools are being used, and people can see, those tools popped up right in their existing workflows. And because they're right in the existing workflows, right, we know that they're delivering, value and and helping to simplify the experience folks are having.
So, with that, I'm gonna hand it back over to to Daniel to to do some demoing. Thank you, Zach. And I'm gonna set a little bit of the stage beyond that. So, again, we're gonna be talking about Canvas next. And, specifically, that's our tier that's very AI forward, essentially.
All of our ignite AI tools are going to be available in that tier. The capstone is definitely the agent, but there are a lot of other Ignite AI features. I'm gonna mention some of those in passing, but they're not gonna be the focus of the demo today. So just kind of setting the stage there. But let's talk a little bit about having your LMS as kind of that center of your AI deployment and what those advantages are.
So Zach's already kind of talked a little bit about this. You have a lot of transparency, and I'm gonna show you once we get into the demo. Pretty much all of our Ignite AI tools have our AI nutrition facts attached to them where you can see what model's in use, you can see how the data is being used, and more. There's configurability. You've got control over what tools you make available to your faculty, to your to your students, so you have that ability to keep things where your comfort level is, where your staff's comfort level is.
Your tools are in context. Right? So our agent, in particular, we're gonna see is aware of what I'm looking at when I do the demo later, and I'm not having to move content between different platforms. I'm not having to go copy my lesson into ChatGPT to get feedback, to copy back over to Canvas. It's all in one place. Right? That agent is going to be that conversational assistance.
We're used to this with LLMs. Nothing crazy new there. But I don't have to worry about technical terms other than maybe a few just key specific Canvas terms just to make sure that things are right, but, again, very conversational. And then educational value, it's really streamlining that workflow, keeping everything in one place, giving you more time to focus on students and help ensure their success. And finally, right, we have many robust integrations in Canvas.
As our agent improves, they're certainly going to be taking advantage of those connections. So that's a lot of talking, and I've got a poll really quick just to get a feel for kind of how y'all are most excited about using AI tools at the moment and what would be most valuable for you. But I promised a little more about me at some point. So I've been at at Instructure for three months now or just about. This week rounds out my third month.
So I'm one of the relative newbies here. But prior to joining Instructure, I spent some time in higher ed, some time in k twelve, some time in instructional technology. And that in that last role, I actually had the opportunity to be a part of our early adopter program for the Ignite AI agent. So I was one of the first people to get my hands on the agent and fell in love immediately, which kind of surprised me. Historically, I've been a little bit of an AI skeptic.
For a while, my stance was kind of really the back and forth here. I'm gonna spend more time getting the the LLM to produce what I want than if I would have just built it myself. But the agent kinda changed my view on that, and the big big piece of that is because it has access to over five hundred API endpoints inside of Canvas. And that's not just pulling data, that's actually executing things and and taking action on my behalf within Canvas. So that's going to include things like changing due dates, creating assignments, creating pages, populating those things with content, sending messages to students, managing differentiation tags, and more.
And that list is growing steadily. Our team has been amazing since I've been a part of this since very early. It's been fascinating watching how that has grown and how it's improved. Things that used to take minutes now take seconds. So that's a lot of talk.
Let's actually get into a demo, and we can go ahead and close that poll as well. I'm curious to see where what y'all are interested in. Alright. So content creation seems to be one of the big ones. We've got some data queries and insights.
Alright. Good news. I've got examples lined up to cover the, the big things here. Now I did mention, there's gonna be a focus on the agent. We do have I do wanna highlight a couple of the other Ignite AI tools.
We have a rubric generator. We have, item authoring assistance for quizzes. We have grading assistance. Right? We have several other dedicated tools. The agent is kind of our bigger, able to handle lots of complex task tools, which we'll eventually be passing off to some of those more specialized tools.
So I'm gonna jump into Canvas and show you a few things here. First and foremost, how easy it is to get to the agent. This lovely button is going to exist at the top of just about every page in Canvas, makes getting there to what I need very easy. I wanna point out one of the very new things our agent team has released is our community library. This is a nice public space, for our users to share prompts that they found are very impactful for themselves with others or to go get inspiration.
So huge kudos for our team for rolling that out recently. I'm gonna get into the AI nutrition facts in a little bit just because I'm gonna need some filler as things are processing, but I'm definitely gonna show you that. And I'm gonna dive in and show you one of my favorite content creation pieces. I'm going to add a disclaimer. This page that I'm about to pull up is going to maybe be a little overwhelming.
It's a wall of text. And my second disclaimer is I teach calculus, and I know people have strong opinions about math. So ignore the content. Focus on kind of what the agent's doing. But this is something I've been doing in my course.
Right? Because I'm still teaching. I've I refuse to let them take that class from me. I enjoy teaching calculus. So here, right, I've provided a transcript of recordings that I've been doing. Right? So I've been recording a lot of content from my course, and it would be amazing to have a text version of that to go along with my video.
Right? Because I know me, I'm old school, give me text anytime over watching a video. I know that's kind of some generational things there. I'm probably showing my age. But it's very easy to come in and ask the agent to give me a lesson. Right? So I'm asking it simply for, hey.
Here's a transcript. Give me a asynchronous learning activity that's aligned to this content. It's gonna take it a minute because it's gotta process what's here, and you can see that's kind of an ugly file. So I'm gonna take this opportunity to highlight those AI nutrition facts. Right? So you can see here the agent currently is based off of Haiku four point five.
If you wanted to see data permissions and those sorts of things, all of that's readily available here as well. Also, I'm gonna add a disclaimer. This is, you know, an AI agent. It may make mistakes. It's it's been acting pretty well.
But every now and then, it decides to go a little crazy. I don't think we're gonna have any issues today. But just to kinda set some expectations, right, I like to think of it as a very enthusiastic TA. Right? It's a great assistant. It does need some oversight.
So you will have to keep an eye on it. You always wanna verify what it's doing. You just saw that lovely wall of text, so it's done. Right? So it has built for me a nice lesson, and the agent is always going to keep me informed. It's going to show me what it's getting ready to do.
It's getting ready to make a page with this content, but it's always human in the driver's seat. And it's always going to ask for my approval before it actually completes any action. So I'm gonna approve this, and it's already got everything done, so it should take it just a moment to make me a page. But while I'm waiting on that, I also have access here to my entire history of chats. So if I needed to come back and resume a chat, continue, iterate on something, I do have that opportunity.
So I've got a direct link to that lovely new thing that it made, so let's go look at it. And from that ugly transcript, it has built a decent lesson, and I can verify that these are exactly the examples that I went over. You know, I probably wanna go clean up the math notation. That's a personal problem, but I would call this probably about a ninety percent good lesson. Now next thing you might be thinking to yourself is, well, that's great, but my institution has a template that my content's supposed to be formatted like.
Or I have a personal look, and I have a personal style that I want this to look like. So I have another lesson that I've built that this is really what I want everything to look like, keep my class nice and consistent, all of my coursework the same. So I can very easily come back to my agent and ask it to create a copy of this and really format it more like that sample. And I'm gonna give it some time to run-in the background. I'm actually not gonna let this one complete because it's gonna take a while, and I've got more things I wanna show y'all.
So I ran this previously, and this is that result. Right? You can see it did a fantastic job of taking that lesson that it built and really making it follow that that blueprint of my own. Now as far as, like, blueprints that you might have at the institution level, right now, this would be something that you would need to have your instructors import in their course, but then it would very easily be very easy to point the agent to that. And this has been I I have not manually edited anything. Right? All of this has been strictly prompting the agent, essentially, take this, produce this.
And you've seen the prompts that I've used to do this. So it's very intuitive. It again, it might occasionally decide to do something a little weird. It's pretty easy to reprompt. Hey.
This doesn't look quite right. Can you try that again? This part's messed up. It's also very easy to conversationally, hey. You added a sixth section here that I really don't want. Can you just remove that piece and make those quick tweaks without having to really dive into the weeds? And this is all well and good.
That's a lesson. Wouldn't it be great if I could do a lot more, though? So, again, staying aware of what you probably already have in Canvas. Right? So, again, context aware, every single one of these prompts, I've said something like this assignment, this page. The agent knows what I'm looking at, where I'm at. Right? So I've got a course outline.
Right? This is one of those things that more than likely you already have in your course anyway. Right? You can see my lovely, slightly complex prompt here asking the agent to take chapter three, which is eleven sections worth of material, and go make a module for my chapter, an assignment for each lesson. I gave it some loose parameters for due dates. I could have added more. I could have told it specifically how many points I wanted things worth, given it more parameters for what to build, but I kept it simple and let it run.
Now this is another one of those that I ran in advance because I wanna show you exactly how thorough it is, and it's going to show me what it's doing in every step of the way. It's going to go create my module. So, again, taking action for me, And then it's going to start building my assignments. And it's going to stop and ask for approval at each step, and then it's gonna go individually move each of those assignments into the module I wanted. Right? So there's a lot of babysitting right now to approve, approve, approve, approve if it's a multistep process.
I will say one of the things that our team is working on to steal a little bit of Olivia's thunder is being able to batch approve this so that I can just approve once for kind of a bulk action once it's given me that game plan to help reduce a little bit of that friction. And using that, it built out my module. And diving into one of those lovely sections, again, is aligned exactly to that template that I wanted. So content creation, the agent is amazing for this. And and I know one of the other pieces to seal again a little bit of oblivious under is going to be eventually the ability to upload and attach files to that that, prompt that you're doing.
So if you have a PDF of your syllabus sitting somewhere, being able to do that instead of necessarily having to have it on a page in Canvas. So that is something that's coming. So that's a lot of content creation. Oh, and I'm trying to keep an eye on the chat. I see a few questions that are gonna be good for later, but I see the one about CityLabs that just came in.
As a former CityLabs user at the institution I was at, right, that was the kind of template I was having to follow, and it can handle that very well. So I can confirm that one. That was actually the use case I was most interested in originally. So that's a lot of design content creation. One of the other things that I think is most impactful here is going to be really helping do some differentiation and making things a little more personal for students.
And and, also, I wanna touch a little bit just on, like, the data and analytics. So here, I have you notice this one's not a fancy one. I haven't cleaned this one up with my template yet. But I have an assignment that some students have done, and I wanna know which students scored under a seventy percent on this assignment. And you might be thinking to yourself, well, I could very easily go to my Gradebook and see that.
Right? It's very easy to go sort by grade, high to low, low to high. There they are. Move on with my life. At least if you've got a small enough class. You got a big class that's gonna take you a little longer, but I've got my students.
But, really, the magic now is what I can do now that I've identified these students. I could create a differentiation tag and put those students in it. Right? Maybe I wanna track them that way. I could do a little more analysis on their current grades in the course. Or what I want to do in this particular case.
And the assignment result here is not going to be amazing because my prompt is very basic, mostly because I wanna keep this quick. But I could ask it to create a remedial version of this assignment. Right? Something that's gonna point out common mistakes, misconceptions, add some additional scaffolding. Right? But I just wanna assign it to those students that I can tell need a little more assistance. Right? I want to be able to ensure their success.
And so it's made that assignment. I gave it some specifications on the amount of points, and it's adding those overrides. So I don't have to go find those students in the list to manually add them to that that exception. Right? The agent's taking care of all of that for me. And now if I also told it to publish and did some more, I could take this a step further.
And let me grab my prompt. And I wanna send a short message to these students, letting them know they've got an opportunity to improve their grades. And, again, the agent is taking action on my behalf. So when it sends this message and I can always come back and review it. Right? Okay.
That looks like it's probably not too bad. Go ahead and send. Or if I don't like it, have some back and forth. Might just say, here's the message I wanna send if I know this is really what I want to send. Right? That message is going to get sent individually to each of those students from me.
Right? The agent is taking action on my behalf, which means those students are going to see it come from me and can reply to me if they have additional questions. So, again, very easy to stay involved, to stay connected, and have that human in the driver's seat. Now from here right. So that's kind of that really bringing you more time to do what matters, and that's have that student impact. This is another of those use cases that I'm most excited about is that whole let's build some differentiated lessons.
I need a version of this that provides additional support. I need a challenge version of this. I've got some high achievers that I wanna kinda push a little further. It really kinda depends on that workflow, but there's a lot of options to really generate a variety of content quickly. Again, TA, always go double check behind it.
Don't just let it run off on its own, but it's there as a great partner. Now the the last thing that I wanna look at is the feedback piece. Right? And this is really good, not just kind of at the instructor level, but this is going to start being something that's valuable kind of administrative level. Right? I imagine a lot of, institutions do some version of, course reviews where you're gonna go review content in a course looking for certain things. The agent is great at that as well.
I'm gonna give kind of a generic set of directions here and ask the agent to just kind of act as a instructional coach for online instruction and give me some some feedback on my assignment, giving me what's working, what's not working, maybe what I can improve. Right? But, obviously, if you had some kind of institution level rubric that you wanted to kinda feed in here, if you followed something like QM or the NIT rubric, you could just reference those by name. And it's gonna give you feedback then on what's on that page based off of what you've asked for. And this should be pretty quick, and this one's generally pretty reliable as well. And there we go.
And so it's given me what's going well. It likes my learning outcomes, my scaffolding. It's pointed out I don't have a rubric. Yeah. I should probably do something about that.
It's pointed out a few kind of Canvas side specific things. Maybe it's not published. Maybe I'm hiding grades. Is that are those settings intentional? Maybe the due date looks a little weird. So it's not just looking at the content on the page.
It's looking at this item in Canvas holistically. Right? And so this is, again, one of those huge advantages of having such a deeply integrated AI in the LMS, right, is it's really not just surface level. It can go a whole lot deeper, and I'm not having to spend time jumping between platforms, copying and pasting, files. Again, really saving time to do what matters the most and improve student impact. So that's really high level, what the agent can do, some of my favorite prompts that I've been using in my course.
And with that, I'm gonna pass it over to Olivia, who's gonna give you all a nice little preview of some of the things that they're working on. Alright. I'm super excited to show you on top of what you've seen from Daniel that's already working in the agent today. One of the many new things that we're working on for the agent that I'm particularly excited about. So let me just share my screen.
Alright. And what I'm gonna show to you today is called prompt builder. As you saw from Daniel's demo, the agent has infinite possibilities. You can ask it to do so many different things, and that's powerful. That makes it very open ended and powerful for many different use cases.
But we also hear from educators that that can be a bit intimidating. It's an empty chat box, and you don't know what you don't know, you don't always know what you want the agent to do or what it can do. So, we've been exploring, in addition to what's already there, having a more scaffolded experience that can help educators with a range of comfort with AI actually get value out of the agent. So, that's what you're seeing here. And this is a a live prototype that you're seeing.
So, this looks very similar to what will actually be in Canvas. And in prompt builder, you can see there are some different actions that I can take. These are all things that the agent can do, and it's gonna walk me through how to build a prompt to do these things. And, right now, there's a list of six things, but this is something that we anticipate we'll be expanding to be adding many more different use cases. So I'll just show you a couple examples.
We'll do scheduling and extensions. This one is quite a powerful use case for the agent. If for teachers who have a class where they need to do, like, bulk date updates or bulk updates of many different assignments or modules. And this this prompt builder flow is just gonna walk me through that. So I can tell it what course I wanna make updates in, and I get some some guidance here on, you know, how I might wanna refer to my course.
If I'm already in a course, it will autofill this. So it'll know what course I'm in and it'll put that in this text box. And then, this is where some of the magic comes in. I I have a a text box here where I can I can type in whatever I want? Do I wanna ship the assignments forward seven days? There there are some examples that I can I can go through here to sort of get my imagination flowing? Do I wanna push all due dates back one week? And, again, I have some some guidance here. So, this is telling me things that I might wanna include in my prompt to make it work better.
So I can say can edit any of these. So we'll just say shift all assignments forward seven days. It'll say all assignments in the class. So, this is not only walking me through it, but it's it's sort of teaching me how to build a prompt and how to tell the agent what I need. And then, I can press this generate Prompt button.
And this gives me my agent ready prompt that's going to do what I need it to do. And when this is live in Canvas, I'll be able to click this Insert Prompt button. And that prompt will go straight to the agent and it'll do the thing. So, it's right there. Or I can copy and paste the prompt if I wanna save it for later.
And I'll just show one more. So, we also have a tags and groups prompt builder flow. This one's a little bit more guided and that just depends on the task. So, for this one, I can choose what kind of do I wanna create differentiation tags or groups? What do I wanna do? Do I wanna create or manage them? And then, again, I have this text box where I can add all those extra details and specific things to my situation that I want to include. And, once again, it will give me my prompt.
And that is Prompt Builder. We're really excited to release this in the near future. And, I will be on the line for Q and A. So, happy to answer any questions that you have about it. Alright.
So we're gonna kick over into q and a. And we've got a few lined up. So I think this one's gonna be for Zach. Is there a way to set up Ignite AI to be a sort of coach or tutor for students in understanding a specific concept and clearing misconceptions, or will the Ignite AI agent only be for instructors? Yeah. That's a great question.
I've seen this a few similar questions here about student functionality. So we we do have a lot of other study tools available for students. The Ignite agent that we showed today is just available for educators, faculty, administration, and staff. Thanks. For Olivia, will the Canvas AI agents be standard across all schools and universities, or will each campus customize which AI tools it has agreements with and only use those? So, the short answer to that question is Ignite AI will provide a consistent foundation, but institutions will have control over what's what gets enabled.
And we are exploring building out MCP integrations within our partner ecosystem. And, as those are built out, you would be able to use the agent and a third party tool together. For example, let's say you use another EdTech product to build out interactive presentations. Once that MCP integration is built out, then you would actually be able to prompt the agent to do that, to to make those presentations for you through that external tool. Awesome.
And another one for you, Olivia. How many tokens would an assignment generation typically take? That's a great question. So, the agent is actually designed to support the vast majority of expected educator workflows without requiring instructors to think in terms of tokens or manage usage at that level. For many common tasks like generating assignments and updating course content, drafting communications, all things that you saw Daniel do with the agent, users shouldn't need to worry about running it to limits. That said, some AI workflows require significantly more compute than others, especially if they're involving very large courses or extensive content or repeated high volume actions.
And, we will have safeguards in place to protect performance cost and responsible use. We're also exploring the right ways to give institutions visibility and control over usage without making the experience feel overly technical for educators. So, this this will evolve over time. And, the goal is to make Ignite AI agent feel simple and useful for day to day work, but still give institutions the oversight that they need. And, Zach, I've got a couple more for you.
There's been a couple of questions around data training and protection of intellectual prop property and privacy. Can you speak a little bit to those concerns? Yeah. Definitely. So the Ignite agent and all other Ignite AI features inside of Canvas are not trained on, institution, or student data. And in in fact, that is, I I think for us, a real feature and not a bug.
It's, you know, we're happy to be able to provide things, that don't rely on that or or don't change the game when we think about student privacy or, protection of of IP. Secondly, I'll I'll say Ignite Agent and, and all of our other Ignite AI features are hosted, using large language models available, through Amazon Web Services, through their Bedrock platform, which means that all of the data that gets sent at prompt time to the model stays inside this the customer's same cloud. Alright. Thank you, Zach. And one more, I think I'm gonna kick your way.
Can each instructor install the Ignite AI agent? Yeah. So, that's another great question. The Ignite agent can be configured on, off, available, unavailable at the account level. Right? So at that kind of top level inside of Canvas. But then it's configurable all the way down through subaccounts and then into courses as well.
Because we know, you know, most folks have got some faculty that are really excited about AI and some faculty that are maybe a little less excited or are not quite far as far along on their journey. So you're able to provide them with with unique experiences. Awesome. Alright. Back to Olivia for a couple more.
Can you speak to, the ability for the agent to be able to take action across multiple courses? Absolutely. Yeah. The agent has access to all of the courses that you have access to. And even when you're not in a particular course, it can see those courses and look at those courses. So, there there are many cross course workflows that we've seen people doing.
For example, if somebody's creating a new course, they might say, this some of this old content from a different course. Pull it into my new course that I'm building. Or searches across multiple courses is another thing that we've seen. But but, yeah, certainly, the agent is able to do that. And then for Olivia as well, I've got a couple of questions around accessibility.
So one one was curious about the accessibility in the prompt builder. And then, also, just some questions around, is the content that the agent is going to produce accessible? Yeah. Great great questions. We know accessibility is top of mind for for many educators. And yes, the agent does default to creating instructional content with accessibility standards in place.
As Daniel has been saying, it's like a TA, you always wanna double check it. But it does typically generate accessible content. And to build on top of that, a common use case we're seeing is actually using the agent to make content more accessible. The agent can look at content in your course and it can scan it, see if there are any accessibility issues, and let you know about those things. And and what you saw in prompt builder is sort of a guided workflow for that.
And and I actually wanna tag in a little bit on that one. I had kind of a surprising use case, essentially asking the agent to do some review for accessibility. And it was a lesson where the instructor had brought in some stuff that was originally done in markup, and the agent actually flagged that there it looked like there were some relics of the markup that hadn't been fixed, which is something that most of kind of our normal accessibility checkers wouldn't have caught. So it's an actual nice extra layer on things sometimes. And for Zach, is there a way for students to know whether certain course components or content is generated with AI? Good question.
So, we we hear a lot of different opinions, from institutions, on this question. So some institutions that, you know, want really clear tags on on everything that's been generated with AI and and others that want to leave it up to the instructor. So, we don't have, the agent kind of tag or stamp its own outputs right now. What we've seen institutions do is is add some standard footer or instruct the agent when it's generating content to make a note about how that content was AI authored. And, Olivia, does the Ignite AI have any sort of institutional guidance or training that'd be taken into account in all of its responses and actions such as, like, brand guides and policies? Great.
Great question. This is a common request request we've heard from from several different institutions that we've been talking to while while developing the agent. And definitely let us know any kind of customization you'd want us you'd wanna see. This is something that we're actively thinking about and exploring and and certainly plan to have some some support for this in the future. And I wanna give a shout out to Olivia's team, by the way.
They're currently hosting regular sessions, I think, right now once a month to get feedback from people who've been getting in and playing. And it's been amazing to see the rapid improvement and and those suggestions that have come into play very quickly. So just I wanna shout out that team for the amazing work they've been doing. And I see one that I'm gonna take for the is this an additional cost or is it available automatically? So right now, the Ignite AI agent is free for US Canvas customers through June thirtieth, and for customers outside of the US through September thirtieth. After that, it's going to be a part of the Canvas next tier, so that'll be all inclusive at that stage.
And for Olivia, how are chats with the agent stored and and how long are they preserved for? Yep. I can take that one. So as as you may have seen in Daniel's demo, we do have that chat history button up on the top of the agent. And, if you click on that, you're able to see all of your chat history, which would be all of the previous conversations that you've had, all of the messages. It looks just like it did in the conversation.
That all is preserved. And, we're having some discussions on data retention. So, we're gonna align with different laws and regulations about how long those are stored. I think, at the moment, it's two years. So, you certainly won't see your chats disappearing for a long time.
And another one for you, Olivia. You mentioned external tools. Will this be able to integrate with textbooks if they're used as an external tool? Oh, that is that is a great question. I was just having a conversation with the team about this earlier today, in fact. We're currently building out our first MCP integrations with the agent with select partners.
And, it's definitely possible that in the future, we'll have integrations built out with text providers. I think that would be very exciting, but not available currently. And for Zach, will the Ignite AI agent be able to identify accessibility issues and fix them? I understand the current Canvas course wide accessibility checker is limited to the Canvas HTML pages. Doesn't check or fix PDFs, will the agent do more with accessibility? Yeah. So I I think my answer is a nondeterministic yes.
Right? I think the the agent knows quite a bit about accessibility. I think it's able to make a lot of fixes, be able to coach and assist on those. You know, but because it's, it's not an on the rails experience like the Canvas accessibility checker, you know, it it may miss things, or it it may overstate some things. But I I think, generally, it does a pretty good job. I I would also say as regards to Canvas accessibility checker, that's another one to keep your eye on because we're, we're always making improvements there as well.
And for Olivia, this one's kind of a follow-up from a previous one, I think. Is the context of the agent limited to a single course at a time, or, is it context for everything you've asked across all of your courses? Great question. So the agent knows what course you're in. So, you know, for example, if I'm on the homepage of my calculus course, it's going to be primarily looking at that calculus course, unless I tell it to look somewhere else. But it does have access to all my courses, and it does like, I don't need to be on the page for it to look at those courses.
So, you know, if I ask, go back to my spring twenty twenty four calculus course or, you know, look across all my courses, it will be able to do that. I think there was also part of the question is, does it have context about everything you've asked? Right now, each conversation is separate. So there's no memory of you know, if you start a new conversation, the agent isn't going to remember what you asked in previous conversations. Although, that is something we've been having discussions around, and we may be adding that kind of memory in the future. And I see one that I'm going to take.
There's a question about sending messages to all students who have under a certain score on a specific assignment. Absolutely. That's one of those workflows that I I showed, like, the very last one I did or one of the last ones I did. You can certainly send messages to assignment based triggers, show me the students who scored under a certain threshold, over a certain threshold. You could also do that as a core score.
Right? I can look at holistic core scores. If I wanted to do a little more diving into some other data points, those are available. Essentially, as long as the agent can identify the set of students, you can target message those students. And I see one more that I'm also gonna tackle, and I think this is gonna be our last question just because we're coming up on time. So I'm gonna go ahead and get our little QR here up as well.
But I just little bit of clarity on, the the agent. I did say earlier that it's free until, June, and there was some questions about deployment. So it requires an admin to enable it. It's a feature flag right now for an admin to enable. So you would need somebody at your institution level if you've got Canvas to go enable that flag.
But, otherwise, that's essentially all that you need to do that for for that free trial period. After the free trial, then that's gonna require a subscription to Canvas Next. And if you want some more information about the tiers, that's what this QR code is gonna get you to. It's some additional information on the tiers of Canvas. And if you're interested more, certainly reach out to, to your instructor rep, and they can get you some more information as well.
So thank you everybody for your time. I think we're pretty much coming on time. It has been an absolute pleasure. Hopefully, that's got you excited for some of the things that we're working on.
Related Content
On-Demand Webinars
-
On-Demand Webinars
-
On-Demand Webinars