Latest / Today’s AI News / Canva's Co-Founder on AI 2.0, What AI Can't Replace in Design, and Who Actually Keeps Him Up at Night

Transcript

Canva's Co-Founder on AI 2.0, What AI Can't Replace in Design, and Who Actually Keeps Him Up at Night · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by Nynex
• 0:03
  Productions. Glad to be here for this one.
• 0:05
  Yeah, it's going to be a good one.
• 0:06
  So you know how using most AI design tools right now is
• 0:10
  basically like ordering from a fast food drive through?
• 0:13
  Oh totally, you just take the bag and pray they didn't mess up
• 0:15
  your. Order right?
• 0:16
  Exactly. You get whatever's in the bag,
• 0:18
  and if it's wrong, well, you just have to order again.
• 0:19
  Start from scratch. Which is super frustrating.
• 0:23
  It really is. But today we're looking at a
• 0:26
  tool that acts a lot more like a sous chef, like someone cooking
• 0:30
  right alongside you, so you can actually tweak the recipe as you
• 0:33
  go. Yeah, a true collaborator.
• 0:35
  Exactly. And we're seeing this play out
• 0:37
  right now with the launch of Canva AI 2 Point O.
• 0:41
  Which is a massive update. Huge Today's deep dive is built
• 0:45
  around an exclusive interview with Canvas Co founder and Chief
• 0:49
  Product Officer Cameron Adams. We're going to unpack how
• 0:53
  they're turning their platform into this fully editable AI
• 0:56
  native environment, and more importantly, what this means for
• 0:59
  the the last mile of creative work, and how your own job might
• 1:03
  actually shift as AI design becomes totally democratized.
• 1:07
  Because the stakes for anyone creating visual content, I mean,
• 1:12
  they have completely shifted with this release, definitely.
• 1:16
  Because if you look at the landscape right now, most AI
• 1:19
  assistants, they stop at the generation phase.
• 1:22
  They hand you a flat file, cross their arms and call it a day.
• 1:25
  Like here's your JPEG. Good luck.
• 1:26
  Exactly. But the bottleneck in any
• 1:29
  creative profession isn't usually getting that first
• 1:32
  draft. The real bottleneck Canva is
• 1:34
  trying to eliminate is the work that comes after that initial
• 1:38
  prompt. A messy part.
• 1:39
  Right. The editing, the tweaking, that
• 1:40
  refinement phase where the real heavy lifting of design actually
• 1:43
  happens. Yeah, I mean, we all know the
• 1:45
  drill by now. You type a text prompt, you
• 1:47
  wait, you know, 10 seconds, and you get this highly polished
• 1:50
  picture. Almost too polished.
• 1:51
  Sometimes, yeah, exactly. But transitioning from that
• 1:53
  static single layer output to a collaborative, fully editable
• 1:58
  design space, that is a massive. Hurdle, it really is, and.
• 2:01
  The interview highlighted that Canva AI 2 Point O is
• 2:05
  fundamentally trying to solve this problem of understanding
• 2:08
  what a user actually means, rather than just what they type.
• 2:12
  Right, getting to the intent. Exactly like take a phrase like
• 2:15
  tension in the negative space for say, a minimal ad campaign.
• 2:20
  Good luck getting a regular AI to understand that.
• 2:22
  Seriously, most text image models simply do not comprehend
• 2:26
  the underlying graphic design theory behind a concept like
• 2:30
  negative space. They just, you know, mash
• 2:32
  together pixels that are statistically associated with
• 2:35
  those keywords in their training data.
• 2:38
  Yeah, and that's because traditional AI systems, they
• 2:40
  learn almost entirely from finished outputs.
• 2:43
  Oh right, like completed images. Exactly, They ingest millions of
• 2:46
  perfectly completed images from across the Internet, but what
• 2:48
  they miss entirely is the actual journey required to get.
• 2:51
  There the process. Right.
• 2:53
  They never see how a human designer actually arrived at
• 2:56
  that final state, so Canva took a radically different route for
• 3:00
  their foundation model. What did they do?
• 3:02
  Well, they trained the Canva design model on tructured data,
• 3:05
  yeah, but also the actual sequence of edits used to build
• 3:10
  millions of designs on their platform.
• 3:12
  Oh wow. Yeah, by leveraging their
• 3:14
  265,000,000 monthly active users, they basically created
• 3:18
  this massive data set built on human hesitation.
• 3:21
  Hesitation. Yeah, hesitations, creative
• 3:24
  pivots, and those like tiny moments of clarity that happened
• 3:27
  during a normal workflow. OK, let's unpack this for a
• 3:29
  second. Because if they are training the
• 3:32
  AI on every single move a user makes, that inherently includes
• 3:36
  every mistake. Right, Absolutely.
• 3:38
  Tons of mistakes. It sounds like, I mean, it's
• 3:40
  like they aren't just teaching the AI to look at a finished,
• 3:43
  beautifully played game of chess, right?
• 3:45
  They're teaching it to study every single strategic pivot and
• 3:48
  every terrible blunder A grandmaster makes to get there.
• 3:51
  But why would a company want an AI learning from user
• 3:54
  hesitations and bad design choices?
• 3:56
  I know it sounds totally counterintuitive at first, yeah,
• 3:59
  but understanding the mistake is exactly how the AI learns the
• 4:03
  user's intent. Like, think about when a human
• 4:06
  designer tries three different fonts.
• 4:08
  They scale a headline way up, decide it looks terrible, scale
• 4:11
  back down, and then finally settle on a specific left to
• 4:15
  line layout. We've all been there.
• 4:16
  Exactly. The AI is learning the context
• 4:20
  of those decisions. It learns why the final choice
• 4:23
  works better than the abandoned choices.
• 4:25
  Oh, that makes sense. Yeah, and that shared thought
• 4:27
  process fundamentally changes how the tool operates.
• 4:31
  The Canva design model doesn't just sit on a server blindly
• 4:34
  generating images, it lives inside the editor.
• 4:37
  So it's right there with you. Exactly.
• 4:39
  It operates within the same typography systems, the same
• 4:42
  layout rules, and the same brand kits you use every single day.
• 4:46
  So when you ask for tension in the negative space, it combines
• 4:50
  the reasoning capabilities of a language model with deep
• 4:53
  structural design training. It actually shares its thought
• 4:56
  process back to you on how it plans to build the individual
• 4:59
  layers. Which I guess brings us to their
• 5:01
  Leonardo dot AI acquisition from a couple of years back.
• 5:04
  Oh yeah, that was a huge piece of the puzzle.
• 5:06
  Because they aren't just giving you like a pre-existing template
• 5:09
  from 2019 and swapping out the text.
• 5:11
  The system is generating entirely new, fully editable
• 5:14
  elements on the fly to fulfill your specific vision.
• 5:17
  Exactly. You aren't getting a flat JPEG
• 5:19
  that forces you to open up Photoshop just to, you know,
• 5:22
  mask out a background or fix a typo.
• 5:24
  You are getting individual layers, scalable vectors and
• 5:28
  live text boxes. In, generating those editable
• 5:30
  elements means you finally escape those frustrating prompt
• 5:34
  loops we've all been trapped in. The worst?
• 5:36
  Right, instead of repeatedly typing, make the logo slightly
• 5:39
  more to the left and just hoping the AI suddenly understands
• 5:43
  spatial reasoning. Which it never does.
• 5:45
  Never. But now the AI simply gives you
• 5:47
  the components. You can just grab your mouse,
• 5:49
  click the logo and physically move it yourself that.
• 5:52
  Is so much easier. It really is.
• 5:53
  It gets you to a usable outcome so much faster because it
• 5:57
  understands the actual anatomy of a design document, not just
• 6:00
  the surface appearance of a picture.
• 6:02
  But wait, if it's generating editable vectors based on
• 6:05
  observing millions of user edits and, you know, hesitations, what
• 6:11
  happens when a user hesitates because they're just making a
• 6:14
  terrible design choice? That's the $1,000,000 question.
• 6:16
  Right, Like how does the AI know the difference between a
• 6:20
  brilliant creative pivot and a genuinely bad layout?
• 6:24
  Because in graphic design, a single misaligned text box or a
• 6:28
  clashing color palette can completely ruin the entire
• 6:31
  piece. So Keva handles this through a
• 6:33
  really fascinating training methodology called Perturbing.
• 6:36
  Perturbing, yeah. They actively stress test their
• 6:39
  AI by deliberately breaking designs.
• 6:41
  Wait. Breaking.
• 6:42
  Them. Yeah, they will take a perfectly
• 6:44
  good, highly effective layout and purposely mess up the
• 6:47
  spacing, or they'll ruin the visual hierarchy or apply
• 6:51
  totally illegible color contrasts to text.
• 6:53
  But doesn't feeding the model broken and ugly designs risk
• 6:56
  like polluting the entire AI? You'd think so, right?
• 6:59
  Yeah, I would imagine it might just learn the bad habits that's
• 7:02
  being shown if you shove enough bad data into it.
• 7:05
  Well, not if you train it to act as a harsh critic rather than
• 7:07
  just a generator. Ah, OK.
• 7:10
  The perturbation isn't designed to teach the AI how to make a
• 7:13
  layout, it's used to train a scoring mechanism.
• 7:16
  The AI is tasked with looking at the broken design and then
• 7:20
  recognizing the specific errors against large scale patterns of
• 7:24
  real world usage. Things like readability and
• 7:27
  alignment. Exactly.
• 7:28
  Readability, Structural alignment, brand consistency.
• 7:31
  It has to score the severity of the mistake and then
• 7:35
  mathematically fix it. They actually call this
• 7:37
  capability agentic editing. OK, we hear the word agentic
• 7:41
  thrown around constantly in tech right now.
• 7:43
  It's like the buzzword of the. Year for sure.
• 7:45
  But what does agentic editing actually look like for the
• 7:48
  person sitting at the keyboard just trying to get their work
• 7:50
  done? So generative systems, they make
• 7:53
  something and then they stop. An agentic system, on the other
• 7:56
  hand, stays awake and acts on your behalf.
• 7:58
  OK, I like that. Stays awake.
• 8:00
  Yeah, it acts as an autonomous agent that continuously scans
• 8:03
  your document while you work. So say you drag a text box
• 8:07
  slightly off center while you're trying to rush through a
• 8:09
  presentation. Which I do all the time.
• 8:11
  We all do. The agent sees that structural
• 8:14
  break in real time, it calculates the proper margin
• 8:17
  based on your company's brand kit, and it gently snaps the
• 8:21
  text back into place or it immediately suggests the fix.
• 8:25
  It is actively working alongside you to ensure the mechanics of
• 8:28
  the design stay intact while you get to experiment with the
• 8:31
  creative direction. Which perfectly explains why
• 8:34
  Cameron Adams is positioning Canva as the critical visual
• 8:38
  layer for massive AI ecosystems like ChatGPT, Claude Copilot,
• 8:44
  and Google Gemini. Exactly.
• 8:45
  It's a very deliberate position. Because all of those chat
• 8:48
  assistants are incredibly powerful for drafting ad
• 8:51
  campaigns or generating, you know, raw concepts.
• 8:54
  But it makes you wonder, are they destined to just be our
• 8:56
  brainstorming buddies? Well, platforms like Canva
• 8:58
  become the actual workshop. Exactly where things actually
• 9:01
  get built. Well, Adams makes a very
• 9:03
  compelling argument about that. He says chat assistants are
• 9:05
  brilliant for thinking, but they are a complete dead end for
• 9:08
  doing. Thinking versus doing.
• 9:11
  I love that. Right.
• 9:13
  Imagine you generate A stellar multi channel campaign idea in a
• 9:17
  chat window. It's great, but the moment you
• 9:20
  need to make a precise change to a visual asset, or collaborate
• 9:24
  with an art director in real time, or ensure the hex codes
• 9:27
  perfectly match your corporate brand guidelines, you just hit a
• 9:30
  brick wall. A text box simply isn't
• 9:33
  engineered for pixel perfect layout adjustments.
• 9:36
  No it's really not. You just end up practically
• 9:39
  begging a chat bot to fix a drop shadow.
• 9:42
  And it never looks right. It's like trying to paint a
• 9:44
  house by passing notes to the painter under the door.
• 9:47
  At a certain point you just want to grab the brush and do it
• 9:49
  yourself. That's a perfect analogy.
• 9:51
  The gap between a text prompt and a publish ready professional
• 9:53
  design is still massive. Yeah, and owning that specific
• 9:58
  space, that last mile of the creative process is canvas
• 10:02
  ultimate Moat. For sure.
• 10:04
  By embedding themselves into the workflows where people do their
• 10:06
  thinking, they become the inevitable destination where the
• 10:09
  actual work gets finished and shipped.
• 10:11
  So if Canva is successfully building that Moat and
• 10:14
  perfecting this last mile, they are essentially making flawless
• 10:19
  execution accessible to anyone. And we really have to talk about
• 10:23
  what that means for the human element of design.
• 10:26
  Because I was looking through the comments on the source
• 10:28
  article and someone literally asked who uses Canva in 2026.
• 10:33
  Yeah, it's a very revealing question, honestly.
• 10:35
  It really is. It shows a lingering perception
• 10:38
  in some circles that the platform is still just a basic
• 10:42
  template library for throwing together quick social media
• 10:46
  posts. Right, like it's just for making
• 10:48
  a quick Instagram story. Exactly This update, however,
• 10:52
  represents A fundamental, massive shift in professional
• 10:55
  capability. Because literally anyone
• 10:57
  marketing coordinator of mid level sales lead a wedding
• 11:00
  planner making a seating chart. Or, you know, a startup founder
• 11:03
  throwing together a pitch deck at midnight.
• 11:05
  Yep, all of them. They can now execute a highly
• 11:08
  polished, mathematically perfect vision.
• 11:10
  And they can do it without waiting in a two week queue for
• 11:13
  an agency or an in house graphic designer to get back to them.
• 11:16
  And that democratization of design is incredibly powerful
• 11:20
  for businesses, but, you know, it naturally triggers some
• 11:23
  anxiety for creative professionals.
• 11:25
  Of course it does. When the gap between an average
• 11:28
  design and a good design effectively disappears because
• 11:31
  the AI guarantees a baseline of quality for everyone, standing
• 11:35
  out become significantly harder. Yeah, if the AI is making
• 11:38
  everyone good, how do you actually be great?
• 11:41
  Exactly. And Adam suggested that when
• 11:44
  execution is fully commoditized, the real differentiator becomes
• 11:49
  the thinking behind the work. The strategy.
• 11:51
  Right. It comes down to human
• 11:52
  judgement, deep empathy, and an instinct for what will actually
• 11:56
  resonate with an audience on an emotional level.
• 11:58
  So it completely reframes the future of creative teams because
• 12:02
  a very common assumption is that putting AI in the loop means
• 12:04
  design departments will just, you know, shrink or disappear
• 12:07
  entirely. Which is a scary thought.
• 12:09
  Very. But Adams flips that premise
• 12:11
  entirely. He argues this isn't about
• 12:12
  shrinking teams, It's about shifting roles.
• 12:15
  Suddenly, every single department in a company has a
• 12:18
  level of design capability they never possessed before.
• 12:21
  Which means the professional edge moves entirely away from
• 12:24
  the raw production of the work. You aren't spending hours of
• 12:28
  your day manually aligning boxes.
• 12:30
  Or finding the right stock photo.
• 12:32
  Right. Or matching hex codes.
• 12:34
  You shift into directing the work.
• 12:36
  You become the creative strategist and the brand steward
• 12:38
  who sets the vision, and the AI just scales that vision across
• 12:42
  the entire organization instantly.
• 12:44
  So the individuals who will thrive in this environment are
• 12:47
  those who can look at the big picture, understand human
• 12:49
  connection, and then guide the AI to execute that overarching
• 12:53
  strategy. Exactly.
• 12:55
  Execution basically becomes a cheap commodity strategy.
• 12:58
  Empathy and taste become premium, highly sought after
• 13:01
  skills. OK.
• 13:02
  So we understand the philosophy here, right, directing versus
• 13:05
  producing, but let's look at how this is playing out in everyday
• 13:08
  practice right now. Yeah, the real world
• 13:10
  applications. Exactly because the interview
• 13:12
  included some fascinating quick hits about how the tool is
• 13:15
  actually surprising its own creators.
• 13:18
  Like Cameron Adams mentioned, he personally uses Canva AI to
• 13:21
  generate cryptic visual clues for his kids scavenger hunts.
• 13:25
  I love that example. It grounds the technology in a
• 13:28
  very practical, relatable reality.
• 13:30
  It totally does. He feeds the AI a list of
• 13:32
  locations around his house and it instantly designs finished,
• 13:36
  beautifully themed printable clues.
• 13:39
  It saves him hours on a weekend project that normally would have
• 13:41
  taken a lot of manual effort. But the professional surprises
• 13:44
  are even Wilder, though Adams mentioned the AI is Florida
• 13:48
  State converting ASCII diagrams into polished designs.
• 13:52
  Oh that blew my mind. For context for you listening,
• 13:55
  ACII diagrams are those clunky bare bones flow charts made
• 13:58
  entirely out of keyboard characters.
• 14:00
  Just dashes, slashes, brackets, IE symbols.
• 14:03
  Exactly. Usually buried in old text files
• 14:05
  or legacy code documentation. So an AI looking at a jumble of
• 14:10
  raw keyboard text, understanding the structural logic of a
• 14:13
  complex flow chart and then instantly converting it into a
• 14:16
  beautiful fully editable graphic.
• 14:19
  It's crazy. It's a massive leap in visual
• 14:21
  comprehension. It shows the model understands
• 14:23
  abstract structure, not just literal images.
• 14:25
  He also talked about using the tool to tidy up messy brainstorm
• 14:30
  slides. And I mean, we've all been in
• 14:32
  those frantic meetings where people are just throwing text
• 14:35
  boxes, sticky notes and screenshots onto a shared slide.
• 14:39
  And it ends U looking like a complete disaster.
• 14:41
  A total mess, but now you can just ask the AI to clean up the
• 14:45
  mess, establish a visual hierarchy, or inject A neatly
• 14:49
  formatted table based on a wall of text you've hastily jotted
• 14:53
  down. And this rapid, highly practical
• 14:55
  utility really explains why Canva isn't losing sleep over
• 14:59
  massive competitors like Adobe or Figma.
• 15:01
  Which is interesting, because you'd think they would be.
• 15:03
  Right. But when asked who kept him up
• 15:05
  at night, Adams pointed out that Canva is playing in a completely
• 15:08
  different, exponentially larger market.
• 15:10
  They aren't just targeting the professionals.
• 15:11
  Exactly. They aren't just targeting
• 15:13
  getting that small percentage of professional designers in the
• 15:15
  world, they're bringing design capability to everyone else.
• 15:19
  They are building a workspace for the non professionals to
• 15:22
  collaborate and turn AI generated concepts into tangible
• 15:26
  real world work. Here's what I struggle with
• 15:28
  though. The user data Adam shared
• 15:31
  suggests people don't want a magic Make it for me button.
• 15:35
  Which is surprising. Very.
• 15:36
  They actively reject full automation and want subjective
• 15:39
  control, but does the average non designer really want to sit
• 15:43
  there tweaking settings? I would think a tired marketing
• 15:46
  coordinator at midnight absolutely wants a magic button
• 15:49
  that just does the job. You would definitely assume
• 15:51
  people want 0 friction, but the data shows users want to be the
• 15:56
  director, not a passive bystander.
• 16:00
  They want the AI to do the heavy lifting of the layout and the
• 16:03
  pixel pushing, but they still want to be able to say make this
• 16:06
  feel more premium or make the tone a bit warmer.
• 16:09
  I see. Yeah, they expect the AI to
• 16:11
  understand the nuance of those subjective emotional adjectives
• 16:15
  and apply them intelligently to the design layers.
• 16:18
  It allows the user to retain final creative control over the
• 16:21
  vibe of the piece. And the data also showed users
• 16:24
  genuinely do not care what underlying model is powering the
• 16:28
  AI. They aren't shopping for the
• 16:30
  latest version of whatever large language model is trending on
• 16:32
  social media that week. Nope, they really aren't.
• 16:35
  They just want to know if the tool fits seamlessly into their
• 16:37
  workflow and helps them get their job done faster.
• 16:40
  The. Underlying technology is fading
• 16:42
  into the background, which is exactly what should happen.
• 16:45
  It allows human choices to come to the forefront, and honestly,
• 16:49
  this ties directly into Adams's most striking insight from the
• 16:52
  entire interview. What was that?
• 16:55
  Well, when he was asked what AI will never do better than a
• 16:58
  human, his answer wasn't about a complex design technique or a
• 17:03
  specific software skill, right? It was about restraint.
• 17:07
  Oh yeah, he pointed out that the one thing AI will never do
• 17:10
  better than a human is deciding not to do something.
• 17:13
  Exactly. AI is an engine of infinite
• 17:15
  options. It can generate 100 variations
• 17:17
  of an idea in seconds, perfectly rendered.
• 17:20
  But it doesn't know which one is actually good.
• 17:21
  Right. What it completely lacks is the
• 17:24
  gut feeling, the lived human intuition to know that 99 of
• 17:28
  those ideas should be discarded. It doesn't know when an idea
• 17:31
  lacks soul, when it's culturally inappropriate, or when it just
• 17:35
  doesn't fit the moment. The ability to look at a sea of
• 17:38
  infinite options and say no. To have the human taste, to
• 17:43
  select the single, quiet idea that actually works.
• 17:46
  That is the ultimate creative skill.
• 17:49
  That completely reframes what it means to be a creator right now.
• 17:51
  I mean, think back to that sous chef we talked about at the very
• 17:54
  beginning. The sous chef can chop all the
• 17:56
  vegetables perfectly and they can manage the temperature of
• 17:59
  the store with mathematical precision.
• 18:01
  But it is still the head chef who tastes the sauce and decides
• 18:04
  if it needs a pinch of salt or if it belongs on the menu at
• 18:06
  all. That's spot on.
• 18:08
  Which leaves you with one final question to ponder.
• 18:11
  If AI is going to handle all the tedious execution, the pixel
• 18:14
  pushing, and the Polish, will the ultimate most highly valued
• 18:18
  creative skill of the future simply be your taste?
• 18:22
  Will your entire professional value come down to your human
• 18:25
  ability to confidently say no? You are the head chef now.
• 18:29
  Make sure you subscribe to the show so you never miss these
• 18:32
  deep dives into the tools shaping our future.
• 18:35
  If you enjoyed this conversation, we'd love it if
• 18:37
  you could take a second to rate the show 5 stars.
• 18:40
  It really helps us out. Keep developing that pace and
• 18:43
  we'll see you back for tomorrow's episode.