Your browser doesn’t support HTML5 audio
Canva yesterday set out its pitch to expand beyond visual design to become the agentic automation platform for all of your work. On the keynote stage at its annual Canva Create conference, the company's three co-founders described their ambition to broaden access to AI capabilities in the same way it has set out to put visual design in the hands of non-specialists. Cliff Obrecht, Co-Founder and COO, says:
We're going to do what we've always done, which is essentially take what is constrained and not used by many and quite difficult, expensive and hard to figure out, and bring that to the masses — make complex things simple. We did this with design many years ago, and we're going to do this with agentic AI — really bring the full power of agentic AI to the people who don't currently access it.
AI makeover
That vision is brought to life with the launch of Canva AI 2.0, a complete makeover of the Canva platform that puts agentic AI at its heart. There are five elements to the new agentic architecture. Some are new, some are adaptations of existing features:
- Conversational design — what was previously the 'Ask Canva' feature is now an integral part of the Canva home page. As an alternative to selecting a specific app or feature to get started, users can just describe what they want to create and the AI platform will figure out how to deliver it.
- Iterative agentic editing — Canva AI can be invoked at any point during the design process. The user can opt to let it decide on elements such as images, text, fonts, colors and positioning, either right from the outset or whenever they want to hand over. Conversely, the user can take back control and make their own edits and changes at any point.
- Memory library — in one of the most significant new features, Canva AI has the ability to learn from a user's or team's design history and adapt to their style and preferences based on this context. In the first iteration, it will build an 'About Me' document based on analyzing all of the user's existing designs in their Canva history. It will then continue to update this 'book' over time based on ongoing activity. The 'About Me' book can also be edited by the user to adjust any elements that need fine-tuning. Canva will be introducing further books over time to build up a complete memory library covering various aspects of user, team, project and brand preferences.
- Layered object intelligence — this is an adaptation of Canva's recently introduced 'Magic Layers' function, based on the vendor's proprietary foundation model which is able to analyze any design and separate it into individually editable layers and elements. This means the AI can change individual elements such as rewriting a headline or changing a background without affecting any other part of the design.
- Intelligent workflows — these are specific sets of actions, or 'skills' in AI lingo, that the AI can invoke to achieve a given outcome or goal. Several of these are existing Canva capabilities, adapted for AI use, such as brand intelligence, where the AI can now automatically apply brand guidelines to whatever it creates, or template remix, which enables discovery and adaptation of design templates. Another example is Sheets AI, which brings the conversational AI interface to Canva's spreadsheet tool, so that users can simply describe the outcome they want and the AI will generate the sheet with the necessary rows, columns and formulae, as well as populating it with relevant data. There's also a new version of the Canva Code vibe coding capability for web pages and widgets, which first launched at last year's Canva Create. This now has the ability to import existing HTML for editing, and publish the finished result back to the user's domain. But several more are worth drilling into in further detail.
Connected workflows
The most significant of the new intelligent workflows connects Canva AI into third-party applications, for example to bring in context and data from Gmail messages, GDrive files and calendar entries, or from Slack threads. The connectors available at launch or soon after also include Atlassian, Figma, HubSpot, Linea, Microsoft 365 and Outlook, Notion and Zoom. Co-Founder Cameron Adams, Chief Product Officer at Canva, explains:
Today, most people have stuff spread out across a ton of different places, and Canva AI connectors enables you to go out to different different sources and incorporate those right into your Canva designs... [For example] you can ask Canva AI to actually go out to your calendar and get it to put together a doc based on the next meeting that you have, and it will produce a fully editable Canva design based off what is in your calendar.
A similar workflow is called web research. This goes out to the Web and collects together research on whatever the user asks for, and then structures and formats it as editable content. Examples include compiling a business proposal, preparing competitive market research, or collating information about a B2B sales prospect.
The above capabilities will often be coupled with scheduling, a workflow for setting up a repeatable agentic task or process that runs automatically at a set time every day, week or month. Examples include compiling a set of briefing documents for meetings on your calendar that day, generating a batch of social content every week automatically formatted for different channels and languages, or creating a monthly team newsletter with all of your latest news collected from Gmail and Slack, ready to edit before sending it out.
Overshadowed by the AI announcements, but much-requested by existing Canva users, a new offline mode was also announced. Available as standard in both desktop and mobile apps, this lets users continue working on an existing design in Canva even when disconnected from the Web, for example when in transit or at other times when connectivity is patchy. Any changes sync automatically once reconnected. Canva also announced a deepened collaboration with Anthropic, embedding access to its visual design capabilities directly into Claude.
Work in Canva
Coming back to the main announcement, it's the ability to connect to other apps that people use in their day-to-day work, as well as out to the Web, that broadens Canva's reach beyond its traditional domain of design and allows it to target becoming "the place where work happens." Adams comments:
I think we're the only scale platform that brings productivity and creativity together, and Canva AI 2.0 is the further expression of that sweet spot in the middle, because we're bringing more of your productivity into Canva. It's already a place that a ton of people do their work in. Docs has been one of our fastest growing design types of all time. We have 100 million people making presentations in Canva every single month. So there's a lot of work being done in Canva already, and having stuff like connectors and web research enables you to bring more of that work into Canva, where you can creatively express it and make amazing visual content that talks about your strategy, about your itineraries and your agenda, brings your team together and enables you to get your message out further.
So we're not only the design platform now, we're an AI platform that's bringing your workflows together in an extremely visual way... Now you can do entire tasks in Canva that literally save you hours of work. Being able to pull in stuff like Slack, Gmail, Calendar context enables you to create designs that you probably wouldn't have tackled before. Those weekly newsletters that you send out to your internal team, [you] probably would have put it in the too hard basket because [it] took up too much time, took up too many people. Now you can do that in the space of minutes or even zero seconds, if you've scheduled it beforehand. It's workflows like that that are really exciting that Canva AI 2.0 is enabling.
Platform rearchitecture
The recent introduction of the Canva Design Model, the vendor's own proprietary foundation model, has been the key to unlocking many of the features of the latest release. It's been part of a broader rearchitecting of the platform and the company as a whole over the past two years that's been necessary to accelerate product development. Obrecht explains:
In order to get the product velocity we now have, we needed to replumb the whole organization and our technical stack. We also needed to set up a really robust research function, so we now have a research team of over 100 researchers — and they're just the hardcore researchers, not the periphery and applied scientists and the delivery folk. And now, once they've got the bedrock of our foundational models built and are focusing on different areas, the velocity of their releases has gone from six months to three months to now every sort of two weeks. And now we're operating at the real pace and velocity of the true frontier AI labs.
Whereas its first generation of models took over two years to develop, he says that research team is now able to train, evaluate and deploy new models in as little as a month, driven by advances in training infrastructure, model architectures, and closed-loop reinforcement learning systems. Adams adds:
The last six months have enabled us to do stuff that we didn't think would be possible for five, maybe 10 years... Back in October, the Canva Design Model enabled you to generate designs in a pretty good fashion, but actually interacting with the designs themselves through AI was still a bit hard. We also launched Ask Canva at the time, so you could ask Canva through a comment to manipulate certain aspects of your design and also give you some advice on your design. But it couldn't really take action on all the elements in a design.
One of the critical things that Canva AI 2.0 does now is it's a full end-to-end creative partner. It's there with you at every stage of the creative process. It's not just one prompt that you start with and then a bunch of manual editing that you have to do. It is there if you start your own design, or if you get it to generate a design when you're halfway through a pitch deck, you can ask it to tweak one slide, you can get it to introduce five slides. You can get it to create multiple social media posts.
It's there at every stage with you, and it intimately understands not just this design, but also all of your Canva context. It knows other docs you've created, other content you've created, and it can bring that to bear as it's working with you in the current design.
My take
Canva's trademark zeal for taking unnecessarily complex technology and simplifying it down to the key essentials shines through in this announcement. That's not to denigrate the underlying technology, which is remarkably advanced and gives Canva a unique proprietary advantage in its core realm of visual design. There's nothing dumbed-down about either the foundation model or the AI capabilities that run on top of it. What Canva has engineered is on a par with what we're seeing from other enterprise application vendors that are leading the charge to reinvent their platforms for agentic AI — the same conversational AI interface, deep orchestration capabilities, a collection of crucial skills and actions, connectors to other apps and knowledge stores, and a strong contextual framework for making sense of the user's intentions and delivering the outcomes they expect. It's just that Canva is making all of that available without its users having to understand any of how it all works under the hood, or having to figure out how to plug it all together.
I was particularly impressed by the memory library, which stores the all-important context that will govern how Canva AI brings together knowledge and actions to deliver the outcomes users ask for. Setting the right context framework is fundamental to getting the best results from generative AI's probabilistic technology, but creating this requires a new approach and skills that differ from those required for the deterministic workflows of conventional software. It's something we're all having to learn. I think it's a genius move for Canva to start this off with a simple 'About Me' document that users can edit for themselves, easing them into this new way of working by intriguing them with curiosity about how they appear in the document. That learning through experience can then grow as the library expands over the coming months and years to encompass other context areas.
Helping users in this way also helps Canva, because the more accurate its context framework becomes for each of its customers and users, the stickier its platform becomes. Canva has already proven in the visual design realm that it can capture enterprise market share by removing barriers to adoption of its design tools while still supporting enterprise governance. If it can pull off the same trick in the agentic AI realm, it will further cement its position. But this is a highly competitive field, with every enterprise vendor — plus Anthropic and other AI-native players — trying to achieve the same contextual lock-in to their own platforms. Succeeding in visual design is one thing. It's quite another proposition to persuade people to put a visual design platform at the center of their work — in preference to, say, a document creation, content storage, team collaboration or business operations platform. On the other hand, who's to say that visual design can't prevail in a digital world that is, after all, becoming increasingly multi-modal? We'll have to wait and see how this all turns out.