Jun 20, 2025 · 56m · mad
Inside a $3B ARR Rocketship — Canva CTO Brendan Humphreys on Deploying AI at Scale
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Canva CTO Brendan Humphreys joins host Matt Turck on The MAD Podcast to discuss Canva reaching $3 billion in ARR, scaling its global engineering team, and deploying flexible AI architectures across its visual communications platform.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 27.1% of the talking time here. How this is scored →
Matt as informed peer 3.3 Guest teaching 3.8 Guest disagreement 0.3 Matt pushing back 0.4
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 25:00 Brendan rejecting the disintermediation threat of conversational agents
When Matt double-clicks on whether external agents mediating user workflows threatens Canva's core platform relevance, Brendan firmly reframes the premise, arguing that conversational interfaces hit clear functional limits in complex visual design.
Hardest push from Matt ▶ 41:38 Matt challenging custom foundation model investmentsMatt directly challenges Canva's strategy of acquiring and training custom models, citing the prevailing tech consensus that general frontier models will surpass domain-specific models.
Biggest teaching moment ▶ 34:42 Brendan on the bipolar impact of AI tools across seniority levelsBrendan provides a nuanced lesson on software engineering development, explaining how AI coding tools create a 'bipolar result' by vastly accelerating senior engineers while creating dangerous blind spots for junior engineers who lack CS fundamentals.
Matt holds his own ▶ 1:16 Matt reciting comprehensive Canva business metricsMatt demonstrates deep background knowledge by precisely reciting Canva's $3B ARR, subscriber metrics, growth rates, customer numbers, and early history before the guest introduces them.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| Celebrating Canva's $3B ARR and Scale Metrics | 4 | 3 | 0 | 0 | Matt demonstrates preparation by rattling off detailed company metrics including $3B ARR, subscriber numbers, and Fortune 500 penetration. Brendan provides early company history and explains Mel Perkins' founding vision. | |
| Engineering Leadership, High Agency, and Scaling the Org | 3 | 4 | 0 | 0 | Matt highlights the high proportion of engineers (2,300 out of 5,000 employees). Brendan shares management insights around founder agency, avoiding the bystander effect, and scaling engineering management. | |
| Managing a Global Distributed Team and Asynchronous Collaboration | 3 | 3 | 0 | 0 | Matt observes the post-COVID pendulum swing back to office culture across tech. Brendan details Canva's async technical documentation stack and autonomous regional hub structure. | |
| AI Threat vs. Opportunity & Ecosystem Distribution | 3 | 4 | 1 | 2 | Matt asks if major AI lab releases create existential fear for Canva. Brendan reframes AI releases from threats into distribution opportunities via Canva's ecosystem app store. | |
| History of AI at Canva: From Recommendations to Leonardo.AI Acquisition | 3 | 4 | 0 | 0 | Matt accurately notes Canva's early ML usage with background removal acquisitions. Brendan outlines their 200 ML engineers and platform abstraction layer over third-party models. | |
| Product Tour: Magic Studio, Magic Grab, Canva Sheets, and Dream Lab | 4 | 4 | 0 | 0 | Matt exhibits familiarity with Canva's product suite (Magic Studio, Dream Lab, Canva Sheets, Canva Code). Brendan explains the technical mechanisms behind Magic Grab and raster image decomposition. | |
| AI Agents, Agentic Workflows, and Model Context Protocol | 4 | 5 | 2 | 2 | Matt presses on whether external AI agents mediating user tasks could disintermediate Canva's front-end experience. Brendan respectfully reframes, asserting that visual design requires a workspace destination beyond conversational chat limits. | |
| Engineering Velocity, Domain Modeling, and AI Skill Democratization | 3 | 4 | 0 | 0 | Matt asks how high shipping velocity is maintained at scale. Brendan explains domain modeling, clean internal API contracts, and the shift from deterministic unit tests to probabilistic evals. | |
| Vibe Coding Realities, Code Ownership, and AI Productivity Metrics | 4 | 4 | 1 | 1 | Matt brings up Brendan's recent LinkedIn post questioning 'vibe coding'. Brendan defines vibe coding nuances, insists engineers must own generated source code, and cites a 30% increase in merged pull requests. | |
| Impact of AI Productivity on Canva's Hiring Strategy | 3 | 4 | 0 | 0 | Matt asks if developer productivity gains alter hiring plans. Brendan explains the bipolar impact of AI tools—supercharging senior engineers while risking junior engineers who lack fundamental critical thinking. | |
| Cultivating Empathy and Productive Collaboration in Large Teams | 2 | 4 | 0 | 0 | Matt asks how engineering empathy works in practice. Brendan breaks down practical communication frameworks like 'yes and', steel manning, and the principle of charity in resolving technical debates. | |
| Build vs. Buy Strategy & Model Abstraction Infrastructure | 4 | 4 | 2 | 2 | Matt presents the industry argument that general foundation models render custom model training obsolete. Brendan pushes back, citing Canva's unique opt-in design data flywheels. | |
| Open-Source Models and Model Serving Infrastructure | 4 | 4 | 0 | 0 | Matt raises specific technical concerns regarding stochastic hallucinations in enterprise brand contexts. Brendan describes decomposing complex workflows into deterministic heuristics combined with targeted LLM calls. | |
| Architectural Lessons from Transitioning PLG to Enterprise | 3 | 4 | 0 | 0 | Matt asks for architectural lessons learned during the PLG to enterprise pivot. Brendan explains the heavy retrofitting tax of region-sharded database storage across hundreds of existing services. | |
| Managing Technical Debt and Re-Architecting Canva's Core Editor | 2 | 4 | 0 | 0 | Matt prompts for stories on managing technical debt. Brendan describes using technical debt pragmatically, highlighting a two-year re-engineering effort to replace Canva's original jQuery editor. | |
| Balancing Feature Velocity with Platform Reliability at Scale | 3 | 3 | 0 | 0 | Matt asks how Canva balances rapid shipping speed with platform stability for 230M monthly users. Brendan explains internal dogfooding and feature flags, while revealing an upcoming integration with Google Veo 3. | |
| The Canva Foundation and Rewarding Content Creators | 4 | 3 | 0 | 0 | Matt notes the founders' equity pledge to charity and Canva's creator fund. Brendan details the $200M Creators Fund established to compensate artists whose content trains AI models. |
Statements from this episode (39)
Canva has roughly 400 engineers across Europe via acquisitions
“So we have about 400 engineers right through Europe through our acquisitions.”
Distributed engineering teams need autonomous product seams to function across time zones
“The key is to find ways for teams in, kind of, incompatible time zones to be autonomous, as autonomous as possible. Usually that means just finding the right seams in the product, so there's real ownership in, in these locations, and that's what we've strived …”
Humphreys: Canva is well-positioned to integrate third-party visual AI models
“Look, we're certainly alert to the rise in visual AI, but we feel we're actually really well positioned. We can Use third party models when they become available, integrate them into our platform very, very quickly, and provide a really rich, cohesive experien…”
Humphreys: Niche AI startups gain mass distribution through Canva integrations
“A lot of niche AI startups are finding is a fantastic distribution channel. So, they've got their niche their niche product, they can come and build a rich integration into Canva, and then suddenly they get distribution, they get a lot of users.”
Canva employs over 200 machine learning engineers
“We have, ah, over 200, ah, machine learning engineers on staff.”
Humphreys: Canva built AI model abstraction layer over OpenAI, Bedrock, and SageMaker
“We've invested enormously in putting a platform abstraction over a lot of the third party AI that we use whether that's kind of like base layer infrastructure like Bedrock or SageMaker, or whether it's third party AI providers like OpenAI, and that allows us t…”
Canva Code went from concept to production in just three months
“Canva code is, is an interesting feature that we've just launched just back in April. That's a feature that allows users to build little intelligent interactive widgets in their designs. That went from kind of first idea through to production in about three mo…”
Canva uses agentic AI flows to handle first-line customer support queries
“Internally, we have agentic solutions deployed, for instance, in our customer support, We're using a lot of agentic solution there, where First line support when a query comes in from a user that goes through an agentic flow and quite often we can handle user …”
Canva built an internal implementation of the Model Context Protocol
“We have our own model context protocol that we've stood up just in internally, ah, and we're certainly talking to the big vendors about, ah, interactions there that are possible.”
Conversational AI interfaces have natural limitations for visual design workflows
“If you are doing any kind of visual communication, then there are going to be natural limits to a conversational interface. And people are going to want to be able to collaborate. They're going to want to be able to control for brand. They're going to want to …”
Dedicated AI engineers will eventually become obsolete like database administrators
“We think that, you know, once upon a time in software teams, you'd have a DBA on the team, but now a DBA is not really a skill anymore because everyone, that the abstractions above databases are such that you don't really need that deep skill unless you're dea…”
AI development requires shifting from deterministic tests to probabilistic evaluations
“I think a lot of software engineers are trained, particularly the more senior engineers are trained in, ah, wanting to write a whole set of unit tests based on a deterministic system. But when you've got a non-deterministic system, you have to kind of move you…”
Canva mandates that developers fully understand and own all AI-generated code
“We have a very important rule that you need to own the output of the tool. That means you have to understand it as if you'd written everything yourself, because ultimately the generated source is the source of truth, and it's going to go, go into the repositor…”
Canva engineers using AI coding tools merge 30% more pull requests
“We're seeing about 30% more PRs merged for, from engineers who are using these tools predominantly.”
AI-assisted code review will inevitably match AI code authoring at scale
“Author with AI superpower will be matched by reviewer with AI superpower. I think that's an inevitable path that will go down at scale.”
Canva is slowing hiring to evaluate the impact of AI tools
“We have slowed down hiring a little bit for a number of reasons. We are looking at, you know, we wanted to just take a pause and understand where the market's going with these AI tools. We are still hiring, but it's not hiring as fast.”
AI coding tools pose serious quality risks when used by junior engineers
“There's a challenge for our industry in that we do see kind of slightly bipolar results for these tools in the hands of a senior engineer who can tell good code from bad code. They're very, very powerful, but for graduate engineers, for junior engineers, ah, t…”
Humphreys: Critical thinking cannot be taught without domain experience
“Critical thinking is something you can't really teach. It's something that you need to build around a domain of knowledge, and the only way you get a domain of knowledge is to work in that domain.”
Canva will raise its hiring bar for junior developers due to AI
“So I think at Canva we'll be focusing on probably just more careful onboarding, more careful peer review of grads. We will be more picky. We will be looking for the grads who bring deep first principles thinking strong, really strong CS fundamentals, but also …”
Humphreys: Empathy is a teachable and vastly underrated engineering skill
“You know empathy in engineering is a vastly underrated skill. And it is a skill. It's not a talent. It's something that you can teach engineers.”
Steel-manning arguments resolves technical differences and builds trust in engineering teams
“Educate people about techniques like steel manning, ah, that is listening to someone else's argument that may be different to yours, ah, and then being able to restate it in its, in the strongest possible terms with genuine intent. You know, one of two things …”
Canva is investing heavily in proprietary AI model development
“We are investing heavily in our own model development.”
Canva possesses billions of unique opt-in design data points
“We have billions of very unique opt-in data points Around design creation, design intent, ingredient selection.”
Canva acquired Leonardo.ai primarily for its AI research capabilities
“We primarily acquired Leo for their research capability, though.”
Humphreys: The AI foundational model race has not been won yet
“On the question of foundational models, I don't think that race has been won.”
Canva's proprietary design data uniquely positions it to build foundation models
“So maybe that advice of kind of like, don't be in the foundation, foundational model Space is good general advice, but we think that we are in a quite unique position in terms of the data that we have to pursue that particular avenue.”
Canva CTO praises inference serving on Google Vertex AI and AWS Bedrock
“I think certainly Vertex or Bedrock really impressive inference serving.”
Canva uses Weights & Biases and an internal custom AI evaluation framework
“We use weights and biases for our evals. And we have our own eval framework that we've built.”
Canva's CTO doubts that AI model hallucinations will ever fully disappear
“The hallucination problem is interesting. It's minimized, but it's still there, and I'm not sure it's going to go away.”
Retrofitting enterprise admin controls and data residency is extremely expensive
“Enterprise customers want sophisticated admin controls. They want sophisticated auth. They want data residency. These kinds of considerations, if you haven't baked them into your product early, they become quite expensive to retrofit.”
Failing to abstract region-based storage early cost Canva a monumental re-architecture
“Very early in Canva's architectural journey, we put a quite good abstraction in over our AWS infrastructure, but I do think back now and wonder, oof, we could have just spent a little bit more time Really abstraction, abstracting region-based storage. It would…”
Canva spent two years re-engineering its core editor without adding features
“We went through a huge re-engineering effort. Two years of, to rebuild this editor without adding really much of any new features.”
Canva co-founder Melanie Perkins aims to build the world's most valuable company
“So Mel, ah, has a two-step plan for Canva, which is to create the world's most valuable company, and then do the most good in the world that we can.”
Software engineering will evolve into orchestrating AI agents for mundane tasks
“I don't think software engineering goes away. I think software engineering changes to actually a much more fun job. Where a lot of the mundane kind of plumbing of software engineering is handled for you, and you're able to orchestrate software with these agent…”
Canva is natively integrating Google's Veo 3 video model
“We are integrating VO three, which is Google's new amazing video generation model natively into Canva.”