It feels as though I say this every year, but this year’s CanvasCon Europe was the biggest yet, with more attendees than ever before. This growth in numbers seems representative of an increasing number of university clients in the UK, as well as stronger and broader EMEA representation. In a UK HE sector that is not terribly blessed with many good conferences focused on edtech, this is now fast becoming one of the biggest and most established, albeit focused on a particular company.
Of all the edtech vendors operating in UK higher education, there are few that are as invested as Instructure in what is essentially event marketing, and in general I think this is to others' detriment. These events, particularly as they grow, can have an important, if unmeasurable, positive impact and send a signal to prospective university clients, who were also represented at the conference.
The theme of this year’s conference was Future-ready Education in the Making, which as a title reminded me of being told off for writing run-on sentences. In reality, though, the one topic that really dominated the conference was AI. Now, whilst that might sound as surprising to some as being told of the Pope’s religious denomination, previous conferences have not felt as focused on it. Only last year, I commented on how limited a focus there seemed to be on AI developments.
I think, in part, this represents greater confidence within the company about what it’s focusing on in AI within the product, and also the fact that IgniteAI has become more of a reality within Canvas. But more on that and other reflections on AI later.
How Instructure responded to its cyberattack
Since the last CanvasCon, Instructure was the victim of a significant hack that caused widespread disruption and inevitable bad publicity for the company. While there might have been a temptation for the company not to say much about this, it was a smart move to address it as one of the first things said from the front. There was acknowledgement of the lessons learnt and the disruption caused to clients and their students.
Being upfront about things that have gone wrong and not trying to sugarcoat them is almost always the best strategy a company can take in these circumstances. Far from being terminal for companies, these kinds of things, if you manage to come out of them and show some honesty, can, counterintuitively, build greater trust and loyalty. For what it’s worth, anecdotally, I’ve not seen first-hand evidence that the cyberattack has significantly damaged the company or generated widespread and strongly-held negative sentiment.
In addition to this, the product session also included some time spent on the ways they have been strengthening security and putting new measures in place to support continuity. These include new learning continuity admin features, such as exports of rosters and gradebooks. In an era where threats are only going to increase as technology enables more sophisticated targeting of companies, and with a precedent set, this painful experience may ultimately help Instructure in the short to medium term.
Although people may point to particular distinctions between VLEs that may make some more or less vulnerable than others, my own view is that companies prone to thinking themselves less susceptible should heed these words, if you think you're standing firm, be careful that you don't fall.
The latest Canvas product developments
The session of greatest importance at these conferences is the one centred on how the product is being developed, and this year was no exception. Not only was the product keynote of great interest, but it was also well handled this year, with the right balance of coverage and depth.
The product developments covered were organised into three areas of priority
Simplify educator workflows
Make learning more personal
Turn information into action
Knowledge Chats, MCP and the evolution of AI in Canvas
Two announcements that caught the eye within that framework were both AI-centred. The first of these was called Knowledge Chats, which is not overly dissimilar to features like AI Conversation in Blackboard. It is essentially an instructor-set, parameter-driven dialogue between a student and an AI chatbot that’s aimed at surfacing student understanding, which can then be gauged by an instructor.
These types of developments come under the umbrella of seeking to address a growing emphasis on focusing more on the process and not simply the product in assessment. Ultimately, it represents the most transferable application of the style of interaction that’s indicative of general-purpose AI tool usage to a VLE.
However, it is also limited to discursive processes and therefore has broader disciplinary limitations when it comes to assessing the process. Frankly, I don’t think we talk about disciplinary breadth and depth enough when considering both the limits and the opportunities that the use of AI may present in technologically mediated assessment.
Arguably, the biggest development, though, was Model Context Protocol (MCP), which, to the uninitiated, is essentially Instructure enabling institutions to connect an agent or LLM of their choice (think Copilot, Claude, ChatGPT) to Canvas. This feature will ultimately enable read and write capabilities for students, administrators and instructors. So, for example, students can ask an external agent or LLM to draw on the content and material within the VLE to develop a study plan.
While this broadly feels like a positive development, I think a significant challenge lies in the question of LLM preference. The LLMs or agents that an institution chooses to enable will not necessarily reflect the tools that students prefer or already use.
I think this is a more significant issue than it might appear, because LLMs and AI agents are becoming increasingly embedded in people's everyday lives. As a general-purpose technology, they get used in a wide range of ways, not just in relation to something narrow and specific like studying. People develop established habits, preferences and ways of working with the tools they choose, and can develop a degree of loyalty to them.
So, ultimately, I expect this feature to be better aligned with, and to better serve, university staff who are used to using an institutionally licensed option than students. It is also, of course, contingent on universities being comfortable enabling this facility for students.
In addition to MCP and Knowledge Chats, it was clear that IgniteAI, Instructure’s centrepiece AI development, is now more mature and embedded in the product. In the product keynote and a later session, there were confident and concrete demonstrations of what IgniteAI offers within the platform. These include discussion summaries and insights, rubric and grading generation, a set of study tools for students and the Ask Your Data feature.
Beyond AI specific developments, a new UI design was previewed, including a new block content editor, both of which are due in early 2027, with additional features geared towards making content editing and layout easier.
The persistent problems with digital credentials and badges
Lastly, there was also some time given to the Parchment part of the wider product family, focussed on digital credentials. This part of the session was very much aligned with the zeitgeist of a singular focus on education’s utility for work.
However, to be frank, I think digital credentials are an area that edtech companies are not remotely getting right, and haven’t for a very long time.
Nowhere is this more apparent than in relation to digital badges, which are actually about 15–20 years old. I know some people get rather excited about them, but the simple fact is that they have never, ever really taken off. Yet rather than spend some proper time thinking deeply about why they've failed, we seem to have spent a number of years confusing the persistence of an idea with evidence that it's a good one. Or, to put it another way, let's do some proper experimentation with digital credentials, without the obsession that they must resemble some sort of digital equivalent of a Scout badge.
Another trend that was evident was what I think is a growing obsession with skills and competency taxonomies linked to job roles. This was apparent through the use of TalentNeuron data with Parchment, which aims to connect the badge issued to a particular skill and, ultimately, to the labour market, or at least the version of the labour market constructed by these kinds of technology platforms.
For me, this is another point at which things begin to fall down. The underlying assumption is that hiring decisions are rather more rational and mechanical than they actually are, and that employers recruit by matching lists of skills to lists of people. The fact is, the skills employers list in job adverts aren't necessarily the same things that ultimately determine who they employ. A job advertisement, which is where much of this data comes from, is a description of what an employer has decided to advertise, not necessarily an accurate representation of what they actually value.
Labour-market data is useful, but I do think we're in danger of confusing what we can measure about the labour market with what actually matters within it. We seem increasingly determined to break people down into ever smaller, measurable units of competency, as though employability were simply a matter of collecting enough of the right ones.
This takes you to some bizarre places and implementations, like the issuing of a badge for critical thinking, as though this were some discrete, context-free competency that you can acquire, demonstrate and then carry around with you for the rest of your life, like a Nectar card for intellectual attainment.
The danger in all of this is that we end up building an incredibly sophisticated system for matching students to job descriptions, rather than helping them become the sort of people employers actually want to hire.
Ironically, it is precisely this lack of genuine critical thinking and creativity that has kept digital credentials in a state of perpetual underachievement. While I saw plenty in relation to Parchment that was on trend, I saw very little to suggest anyone was fundamentally questioning the assumptions underpinning the whole endeavour, or doing anything to end digital credentials' long career as a perennial underachiever. A career so long that I reckon they're now eligible for a gold watch.
Overall reflections and the gap between AI developments and educational value
Overall, I think this event was a positive one for Instructure. The company appears to have come out of the hacking incident relatively unscathed and has greater clarity and better messaging about how AI changes and develops Canvas. Outside of these events, it has continued to grow its number of university clients, and I did not witness anything that made me think it is unlikely to continue to do so.
However, I have my question marks over AI product developments, not just within Canvas but within the VLE market as a whole. One of the fundamental problems is that, over the last few years, almost every question being asked has been viewed narrowly through the lens of AI. Companies are not asking where they can bring value or what can make a difference to instructors and students. Instead, it’s all been largely in relation to AI.
Ultimately, what that does is narrow the opportunity space massively.
So, for example, because AI can effectively summarise lots of text, we get discussion summaries, a feature developed from an efficiency mindset, not a value-based one. What are the value of discussions for an instructor? It can be in getting to know students, as much as it is in gauging that 66.7% of students have misunderstood something. It can also be through sending a signal to students through their personal, authentic engagement with them in that space.
In a similar vein, we have AI marking, almost always framed as something that is provided upfront for a human to verify and edit. However, invariably, when you make something easy for humans, they choose the path of least resistance, so my prediction is that we will see more AI feedback and less human editing of it.
Equally, what you get with AI feedback developments is a kind of additive bias, where the ability of AI to create more feedback is equated with that being an inherent improvement. For what it’s worth, I think audio feedback from an instructor creates much greater value than anything AI has currently done in relation to feedback.
Lastly, there is the question of what AI can do with data from a VLE. Instructure’s Ask Your Data is a great feature, insofar as it allows you to work with data through natural language commands. But what AI, and its ability to handle large amounts of data, often gives you is a world viewed almost entirely through the lens of averages. Which is like only experiencing the world in one colour.
So, while the defining theme of this conference was AI, and Instructure seems settled on its role within Canvas in the short to medium term, I remain unconvinced that this multi-year period of AI product developments within VLEs has brought a level of value that matches all the effort and the deafening noise.
