If you only saw the Canva Create 2026 keynote highlight reel, you'd be forgiven for thinking this was just another "type a prompt, get a design" demo. Prettier slides, snappier stock footage, a CEO on stage with the obligatory "this is the most ambitious launch since we started in 2013" line. Move along.
Don't move along.
What Canva announced on April 16 in Los Angeles is the first concrete attempt by a horizontal productivity tool outside the Microsoft/Google orbit to turn itself into an agentic workflow platform. Canva AI 2.0 ships with conversational design, agentic orchestration across its entire tool surface, a memory layer that persists across sessions, and — the part most of the coverage missed — native connectors to Slack, Gmail, Google Drive, Google Calendar, Notion, Zoom, and HubSpot.
That last sentence is not a design feature. That is an operating system for marketing work. And if you run enterprise AI, you should be paying attention to what Canva did here, because the pattern is going to repeat — in sales tools, in HR tools, in finance tools, in every SaaS your company already pays for.
Let me unpack what's actually in the announcement, why it matters strategically, and what I'm taking away for how we plan AI investments at Zscaler.
What Actually Shipped
Canva's wording is marketing-heavy, but the architecture underneath is consistent with what the leading edge of agentic platforms looks like in April 2026. The release breaks into four foundational capabilities plus six workflows.
The four capabilities:
- Conversational design. Describe an idea in natural language, get a fully editable design with hierarchy and brand already applied. Not a template picker. A layout generator that produces editable, layered objects instead of flat images.
- Agentic orchestration. This is the load-bearing piece. The Canva AI assistant has access to Canva's full internal tool surface — layout engines, background removers, image generators, text tools, brand kits — and picks the right tool for a user's intent. A request like "make a 4-slide deck for our Q2 product launch, pull the positioning from our Notion doc, and export assets sized for LinkedIn and our website" becomes a single workflow. Not a prompt → image → manual cleanup loop. An agent.
- Layered object intelligence. Everything the agent produces is a first-class Canva object — editable, groupable, re-styleable. This matters because a flat AI image is a dead end; an editable file is a starting point for a team.
- Memory (the "Memory Library"). User preferences, brand styles, prior design decisions persist across sessions. This is the feature every enterprise AI leader has been asking for from every agentic platform, and Canva shipped it at consumer scale with a research preview to a million users.
The six workflows bolted on top:
- Connectors — Slack, Notion, Zoom, Gmail, Google Drive, Google Calendar, HubSpot (more coming). This is where the platform shift lives.
- Scheduling — agent-driven calendar coordination tied to design deliverables.
- Web research — the agent can browse, cite, and pull data into designs.
- Brand intelligence — the agent learns and enforces brand guidelines without a human copy-checking every output.
- Sheets AI — spreadsheet automation that plugs into the design workflow (think: auto-generated charts with brand styling).
- Canva Code 2.0 — code blocks and embedded interactive elements.
Canva says its proprietary models are 7x faster and 30x cheaper than "comparable frontier alternatives." Take that with a shaker of salt — it's vendor-benchmarked — but the direction is right. Cost curves are collapsing on inference-heavy agentic workflows, and that's what makes the economics of an "agent-per-task" architecture actually pencil out.
The Strategic Move That's Hiding in Plain Sight
Here is the part the design-focused coverage missed. Read the connector list again: Slack, Gmail, Google Calendar, Notion, Zoom, HubSpot.
Those are not design tools. Those are the tools marketing, sales, and customer-facing teams already live in every day. Canva is not trying to compete with Figma anymore. Canva is trying to compete with Microsoft 365 Copilot and Google Workspace Gemini — for the creative-adjacent portion of enterprise work.
The play is: the agent that spans your workplace tools wins the workflow.
Microsoft is running this play from the productivity side with Copilot. Google is running it from the search/docs side with Gemini. Salesforce is running it from the CRM side with Einstein. OpenAI is running it from the raw-intelligence side with GPT agents. Canva is running it from the content-and-brand side — and no one had that lane locked down.
What makes Canva's angle defensible, if they execute:
- They already own the content layer. Canva has 240M+ monthly active users who create brand assets, decks, social posts, videos, and documents in Canva every day. The agent can read existing brand context and produce on-brand work faster than a general-purpose agent that has to be trained on your brand every time.
- Memory + brand intelligence is a retention mechanism. Once the agent has learned your company's tone, palette, photography style, and messaging hierarchy, switching costs rise sharply. That's moat.
- The connector list is the ecosystem, not the product. Connectors mean the agent pulls context from where work actually happens (Slack conversations, email threads, Zoom calls) and outputs into the channel where it needs to land. The design app is merely the surface; the workflow is the product.
If you're a CMO or a head of marketing operations, pause on this. The top-of-funnel content assembly process in most enterprises today involves a dozen tools, three cycles of review, and roughly 40% of a marketing manager's time. An agent that spans Slack → Zoom notes → brand kit → HubSpot campaign with a memory of what worked last quarter is not a productivity improvement. It's the collapse of an entire workflow stratum.
Why This Is a Watershed Moment for Horizontal Agentic SaaS
Zoom out. Canva is the first major horizontal SaaS vendor outside the hyperscalers to ship a production-grade agentic orchestration layer with native enterprise connectors, cross-session memory, and tool-level governance (they're explicit about SSO, SCIM, and admin controls in the Enterprise tier).
Every major horizontal SaaS — Atlassian, ServiceNow, Asana, Monday, HubSpot itself, Miro, Figma, Airtable — is sitting on some version of the same opportunity. They all have:
- A proprietary data substrate (tickets, CRM records, project boards, whiteboards, design files).
- An existing user base with workflow muscle memory.
- Pressure from above (Microsoft/Google/Salesforce building agents that reach into these tools via APIs).
The strategic question every one of these vendors is now answering: do we become an agent endpoint that hyperscaler AIs call into, or do we become an agent origin that orchestrates across other tools? Canva just picked "origin." I expect Figma, HubSpot, ServiceNow, and Atlassian to pick "origin" too, within 180 days. Everyone else will end up as endpoints — valuable, but with weaker margin structures and less durable customer relationships.
For enterprise AI leaders, this means the vendor landscape is about to fragment into two camps, and the org needs explicit policy on both:
- Origin agents (Canva, Copilot, Gemini, ChatGPT Enterprise, Salesforce Einstein, etc.) that orchestrate across tools. These need governance, identity, DLP, and audit logging at the agent level, not just the user level.
- Endpoint tools that origin agents call into via connectors. These need tight scope — what can the agent read, what can it write, what does it expose through its API. The MCP ecosystem is where most of this standardization is happening.
If you're still thinking about "approving AI vendors" the way you thought about it in 2024 — one model at a time, one tool at a time — you're in for a rude surprise. The unit of approval now is the agent and its connector graph, not the tool.
The Governance Problem This Creates
Here's the part that will make CISOs nervous, and rightly so.
When a Canva agent connects to Slack, Gmail, Zoom, Notion, and HubSpot on behalf of a marketing user, that agent can read data in every one of those systems. It can combine them in ways no human would have the patience to. It can output that combination into designs that get shipped externally in minutes.
The obvious risks:
- Data leakage across connectors. An agent that can read a private HR Slack channel and also write a public marketing asset is a bad prompt away from a resume-generating incident.
- Prompt injection through inbound content. Zoom transcripts and Gmail messages are user-generated content. An adversary who gets you to paste their email into a Canva AI prompt has a path to instruct the agent.
- Over-scoped OAuth grants. Most users will accept the default connector permissions without reading them. "Read all Gmail" and "Read all Drive" are extremely broad scopes to hand to an agent with autonomy.
- Audit trail gaps. Each connector has its own logging model. Stitching a single agent action into a coherent audit trace across five systems is non-trivial and mostly not done yet.
This is exactly where the Zscaler AI Guard story gets interesting — an inline agent governance layer that sees the prompts, the tool calls, and the responses across every agent your workforce uses. Canva's announcement is a free advertisement for why that category matters. The more agentic origins the market spins up, the more enterprises need a single plane where policy, DLP, and audit apply regardless of which vendor's agent made the call. I won't belabor the commercial here — I'm biased — but the trend is the trend whether you buy our product or someone else's.
What I'm Telling My Team on Monday
A few concrete takeaways I'm writing up for the AI engineering group this week:
1. Add Canva to our "origin agent" watchlist. We already track Copilot, Gemini, ChatGPT Enterprise, and Claude for Work. Canva AI 2.0 deserves the same level of scrutiny. The marketing and brand teams are going to want it, fast.
2. Revisit our connector-level policy model. Our current DLP rules were designed for a world where a user pasted text into an app. In an agent world, the app fetches text from other apps on the user's behalf. The threat model is different. We need policies that understand which agent is making which tool call, not just "user accessed Gmail."
3. Treat memory as governed infrastructure. Every agent that persists memory across sessions is creating a durable secondary data store about user behavior, brand, and content. That store is now in scope for DSRs, for right-to-be-forgotten requests, for breach disclosure. If you didn't have memory in your data inventory a year ago, you do now — add it.
4. Map the connector graph, not just the vendor list. For each agent our users touch, I want a single-page diagram of what that agent can read and write, at what scope. Visualize the blast radius before you try to govern it.
5. Push for MCP alignment in procurement. When we sign vendor agreements in 2026, I want connector behavior specified in terms of the Model Context Protocol — what tools are exposed, what scopes, what auth model. Vendors that can't describe their agent's behavior in those terms are not ready for enterprise deployment.
The Takeaway
Canva AI 2.0 is not a design update. It is a declaration that horizontal SaaS vendors with proprietary content graphs intend to become agentic workflow platforms. That has real consequences for every enterprise AI leader — not because Canva is special, but because the next 10 vendors in your stack will make the same move, and they will not all execute it as well.
The right response is to stop evaluating "AI tools" vendor-by-vendor and start evaluating agent ecosystems — origin agents, endpoint tools, connector graphs, memory stores, and the governance plane that spans them. Your risk committee's vocabulary needs to change this quarter.
The agentic AI market just got a lot bigger, and a lot messier. Get ahead of the mess.
Rajesh Beri is Head of AI Engineering at Zscaler, where he leads the team building AI infrastructure across sales, marketing, finance, customer support, HR, and security. Views are his own.
Want to calculate your own AI ROI? Try our AI ROI Calculator — takes 60 seconds and shows projected savings, payback period, and 3-year ROI.
