Overview of Decoder — "Canva's CEO on its big pivot to AI enterprise software"
Nilay Patel interviews Melanie Perkins, founder & CEO of Canva, about the company’s major AI push: Canva AI 2.0. The conversation covers what the product actually does, how it’s architected, how Canva is thinking about models, costs and pricing, enterprise opportunity and competition (especially big platforms and Adobe), internal org changes to ship the product, and how Canva plans to manage user trust and creator concerns as AI becomes central to the product.
Key takeaways
- Canva is reframing itself from “a design platform with AI tools” to “an AI platform with design tools.” The emphasis is on turning ideas ("concepts") into editable, layered Canva files that users can iterate on.
- Canva AI 2.0 produces layered, editable Canva files (not opaque images) and orchestrates existing Canva tools (background remover, templates, elements) and connectors (Slack, Gmail, calendars) to build end-to-end deliverables like presentations and marketing assets.
- The product supports conversational/iterative workflows (agentic orchestration) rather than one-shot generation; AI outputs are treated as first drafts for human refinement.
- Canva combines external model partners (OpenAI, Anthropic, Google) with a growing internal research team (~100 people) and its own models (e.g., Magic Layers), aiming to control cost, latency and capabilities in domain-specific visual AI.
- Enterprise is a major growth vector: Canva reports fast enterprise growth (roughly doubled year-over-year), deep footprints in Fortune 500 companies, and sees AI as a way to automate repetitive work and centralize disparate company data into usable design outcomes.
- Pricing approach: tiered AI credits (free, Pro, Business, Enterprise); a promotional “AI pass” (equivalent to $100/mo) was granted to the first one million users to test the system.
- Melanie emphasizes accessibility and control: AI is optional (new AI tab in the editor), outputs are editable, and Canva will keep some products free (she explicitly promised Affinity will stay free).
- Risks remain: public skepticism of AI, potential job disruption, content “slop,” and competitive pressure/self-preferencing from platforms (Meta, Google, TikTok/YouTube) that could bundle lower-cost models or restrict integrations.
Product & UX: what Canva AI 2.0 actually does
- Three-tier framing: pixels → objects → concepts. AI adds a new concept layer: "tell it what you want" and get a draft design.
- Outputs are native, layered Canva files so users can drag, drop, edit elements, collaborate and refine.
- Conversational editing: a new Canva AI tab inside the editor lets you generate drafts by dictation/typing and then switch seamlessly to the conventional Canva editor for manual edits.
- Orchestration: AI can call and combine Canva features (e.g., background remover) rather than leaving users to invoke each tool manually.
- Connectors: Canva AI can pull context from company sources (Slack, email, calendar, documents) to produce work tailored to organizational data and history.
- Agentic/iterative capability: the system is designed for multi-step refinement—not just one-shot outputs.
Architecture & engineering approach
- Canva credits a decade of investment in an interoperable, layered design format that spans presentations, docs, videos and whiteboards; this makes programmatic manipulation feasible.
- Internally described as an “orchestra”: many components and tools must coordinate to create a layered, editable file.
- Technical breakthroughs (culminating in a key change last October) plus a design-focused foundational model enabled the push to AI-driven creation.
- Canva invests both in integrating best-of-breed external models and training internal models for domain-specific features (e.g., Magic Layers) to reduce cost and improve latency.
Models, tokens, costs & pricing
- Model partners: OpenAI and Anthropic are named partners; Google and others referenced; plus Canva’s internal models/research.
- Token/credit model: Canva uses tiered AI credits across plans (free/pro/business/enterprise). The company gave a limited-time AI pass (~$100/month value) to the first million users to experiment freely.
- Canva is investing in its own models where it can add unique design value and reduce per-query cost; LLM query costs have fallen substantially historically, according to Melanie.
Enterprise strategy & value proposition
- Canva positions itself as the central place where design, collaboration, and company knowledge converge—useful for automating recurring work like presentations, campaigns and localized marketing.
- The pitch to enterprises: reduce manual busywork by letting AI ingest internal data and produce finished or near-finished assets in a single place.
- Growth signals: Canva reports rapid enterprise growth, deep Fortune 500 penetration, and broader adoption across teams.
- Security and permissions remain an important enterprise concern (agent permissions, connectors), a topic raised via comparison to Okta’s position on managing agent access.
Competition, market dynamics & strategic risks
- Competitors and pressure:
- Adobe is the obvious incumbent in professional creative tools; Canva sees itself addressing more consumer/SMB/workflow-first needs but recognizes overlap.
- Platform owners (Meta, Google/YouTube, TikTok) could build native creative + delivery stacks and might self-preference or underprice model access—creating a risk for third-party tooling.
- Canva’s defense: being the interoperable place where content across formats and platforms is created, stored and iterated; the ability to publish to multiple platforms and centralize creative workflows.
- Pricing and access risk: model providers/platforms may monetize database or model access; Canva hedges with internal model investment and partnerships.
Company structure, decision-making & product org
- Org model: “cupcake” metaphor — large centralized product/core with many local/empowered teams building on top (small teams, goal-oriented).
- Decision process: goal-oriented teams, weekly show-and-tells and the “Canva Jigsaw” approach to break goals into independent pieces. After a breakthrough, effort scaled from a small R&D group to hundreds company-wide.
- Product development has become “AI-native”: tooling, QA, processes and upskilling across disciplines (designers, PMs, engineers) shifted to embrace AI capabilities.
User & social considerations
- Melanie acknowledges public skepticism of AI; Canva’s approach is product design choices (AI tab, editable outputs, opt-in) to reduce fear of loss-of-control and “slop.”
- Education: Canva emphasizes free access and specific education tools (Learn Grid) to help teachers/students produce curriculum-aligned content—positioned as an equity play.
- Community-first stance: Canva leans heavily on user testing and community feedback (over 1M wishes/year, granting product “wishes”) to shape rollouts and minimize backlash.
Notable quotes
- “We’re moving from a design platform with AI tools to an AI platform with design tools.”
- “Design is to mark an idea.”
- “One-shot generation is sort of like AI 1.0… being able to do iterative agentic orchestration is really 2.0.”
- “AI should accelerate your vision and creativity, not override it.”
- “From a user standpoint, they just get to say what they want and then we go and do the hard work.”
What to watch next (actionable signals)
- Beta expansion: Canva AI 2.0 roll-out beyond the initial research preview / the first-million AI pass—watch for broader availability and usage metrics.
- Pricing & token economics: whether model partners start charging materially more per token or platforms create tolls for database/connectivity; monitor Canva’s internal model deployment and cost improvements.
- Platform competition: whether Meta/Google/TikTok push integrated creative + ad suites that edge out third‑party tools or self-preference content.
- Enterprise adoption & security: enterprise uptake rates, connector adoption, and how permission/kickoff/agent governance evolves.
- Product signals: usage of features like Magic Layers (8M uses in 4 weeks per Canva), adoption of the AI tab, and community sentiment about AI outputs vs. human-crafted work.
Bottom line
Canva’s AI bet is designed to make creation faster and more integrated—turning disparate company data into editable, final-form design assets within a familiar editor. The company’s advantages are a decade of format/interoperability work, a large community, and domain-focused model development. Major open questions remain around token economics, platform competition and public sentiment toward AI; Canva is trying to mitigate those with editable outputs, opt-in UX, enterprise tooling, and internal model investments.