👋 Howdy to the 2,873 new legends who joined since our last edition! You are now part of a 527,698-strong tribe outperforming the competition together.

LATEST POSTS 📚

If you’re new, not yet a subscriber, or just plain missed it, here are some of our recent editions.

🍒 The Perfect Onboarding In 7 Simple Steps. How to set your new members up for success from day one.
💪🏻 Metalab: The Agency That Helped Build the Internet. An interview with Luke Des Cotes, CEO at Metalab.
⚡️ The Fintech King & BILL’s 20-Year Race To $1 Trillion. Serving small businesses all the way to 1% of GDP.

PARTNERS 💫

ACH remains the backbone of recurring and subscription payments—but outdated systems create friction, failed transactions, and unnecessary costs. See how payment leaders:

• Simplify onboarding and reduce drop-off

• Lower operational costs and boost margins

• Build more reliable, modern payment systems

Learn how modern ACH helps drive up to 15% higher conversion and 40% fewer returns.

Every newsletter wants to grow. Few actually do. beehiiv is the platform built by ex-Morning Brew folks who got tired of duct-taping ten tools together, and newsletters running on it grow 2.75x faster. Referrals, recommendations, analytics that mean something, all in one spot. It's the closest thing to a cheat code we've seen for getting your words in front of more people.

Get 30% off your first three months on beehiiv with the code: BILL30

_{Interested in sponsoring these emails? See our partnership options here.}

HOUSEKEEPING 📨

I’m not gonna lie: South Korea is pretty awesome. The food is ace (bulgogi bibimbaps have been my fave), the society is advanced, and there is a heck of a lot of history around this place. Okay, yes, everywhere has a history, but the Korean War is particularly interesting. I feel like it’s the war you don’t really hear so much about.

What was very eye-opening was our trip to the DMZ (demilitarized zone) between North and South, a place where we were lucky enough to sit and chat with a North Korean defector. It’s funny that we sit here reading 1984, talking about how things might be looking more Orwellian than ever before, all while North Korea is a walking, talking Orwellian nightmare. Just makes you think how lucky we are.

INTERVIEW 🎙️

Danny Wu, Head of AI Products at Canva

Danny Wu is the Head of AI Products at Canva, the design platform that hit a $42B valuation in its August 2025 employee share sale and now serves more than 265 million monthly active users. He has spent close to a decade at the company, moving up through product management and product lead roles before taking over its AI work. Today, he focuses on Canva AI, the conversational, agentic layer the company calls its “2.0 era,” announced at Canva Create 2026, with more updates expected soon. His job is to turn a pile of separate AI features, like image generation, background removal, and Magic Design, into one product that can act on your behalf across your entire workflow.

What makes Danny worth listening to is the shape of his role. He does not run a single AI team. He sits across Canva's supergroups as the connective tissue, stitching dozens of teams with shared dependencies into one coherent AI product for hundreds of millions of users. He is also a former freelance graphic designer who now trains models, which gives him a sharp view on the question everyone in design is circling: does AI flatten craft into a single tool call, or does it sit beside the human and make the work better? His answer, and the way Canva builds toward it, is the interesting part. He even has a contrarian habit for staying sharp—days when he uses no AI at all.

What does Canva AI unlock for its users?

Canva has obviously had many AI tools and features throughout our history, including image generation, background removal, and even creating presentations with Magic Design. But the biggest shift, and the biggest unlock of Canva AI, is that it really brings Canva into the conversational, agentic AI era.

Canva AI actually expands the scope of what Canva can help you do. Design is obviously one of our core focuses and what it's optimized for, but it closes a lot of user needs that we simply couldn’t without this agentic AI platform. Things like being able to pull all the context from all the places where your work already happens, like your emails and other communication channels, and having these as inputs that now help fuel your work.

We’re also expanding where and how design happens. Canva Code 2.0 turns vibe coding into a design process, where Canva does most of the heavy lifting to help anyone go from prompt to a responsive app or another interactive experience. Users can also edit the output to truly match their brand via prompts or in the Canva editor like any other Canva design. Our latest partnerships are also bringing design directly into new surfaces like AI Mode in Google Search and Alexa+.

Do you see design driven by words rather than direct manipulation as the future?

I think to some extent, design has always originated from something on a higher level. With AI, this might look like more words and higher-level directions, while staying a human craft during the design process, versus direct manipulation. I don't think that's something to be afraid of, necessarily. For some context, before I became a PM and an engineer, I actually used to be a freelance graphic design artist, so I've spent a lot of time in the tools. There are definitely things I’ve always enjoyed about the creative work path, and a lot of manual things I don't enjoy, and this kind of abstraction is getting moved higher and higher.

That's why one of our core goals and constraints with building Canva AI has been really preserving the smoothest human editing capabilities, so that what you create can live side by side. They're not just images that you then have to re-prompt to edit. They create actual designs, so you can use the hand-editing tools, and they work well together. I think both are powerful and empowering ways.

How do you run your weeks at Canva on the AI side?

My typical week varies. One week could involve deep-diving into new models or new products and features we're working on, trying to get them into a good spot. Usually, that translates into a lot of deep-dive sessions, internal testing, and getting all the different streams that are important for a launch aligned together.

But some of my most exciting weeks are the ones where we get to experiment for the sake of curiosity and growth. For example, we recently had a company-wide AI Discovery Week and hackathon. It was a whole entire week of free-form exploring ideas, trying concepts that might sound good but haven't really been teased out yet. That variety, and those kinds of weeks, are the ones I really enjoy.

What does your team look like inside the organization?

I actually have a pretty unique role at Canva. I'm not officially part of any specific AI team or group. At the highest level, Canva is divided into a number of different supergroups. For example, we've got Teams & Education, which, as the name suggests, focuses on everything related to Canva enterprise, business, and our education and learning experience. Then we've got Generative AI as a supergroup that owns tools like Canva AI.

My role is essentially being the glue, connecting the different teams and connecting the dots. One thing you can do very easily, especially with AI, where the cost of building something and prototyping is lower than ever, is ship your org structure. So much of my role is to do my best to empower the different teams and create a unified AI product that supports all our users, our community, and our needs. Internally, there's quite a degree of flexibility in how teams, planning, and guidance work. But especially in the AI space, most teams have moved to a six-week sprint structure, and we've found this has increased our productivity.

What does the goal-setting look like? And how do you decide what’s worth building?

The way we work, and at least how it works well, is that the six-week sprint cadence is not about limiting ambition, either upwards or downwards. It's really focused on how we operationalize the general coordination problem of having a lot of teams working on AI, which often has very shared areas, shared ownership, and shared dependencies, and still have a fast and informal process, but with enough structure. So if you're someone at Canva working on how to handle user uploads, images, and videos at a large scale, which is going to be expanded with AI video features, you do have a little bit of confidence. Goals should be as ambitious as you want them to be.

One thing that's been fairly core to our DNA, even before AI, is that we've had a really low bar for ideas and experimentation. Before AI made prototyping super accessible and empowering for everyone, one of the things we've always done is that if you had an idea for something, the first step was to throw together a little wireframe. You put together some mocks of how you imagine your idea will look, circulate them among a few folks, see if they gain a little bit of traction and excitement, and maybe prototype it.

A lot of it is actually very bottoms-up. Someone says, ‘Hey, I've extended Canva AI with these capabilities, and now I can do all of these tasks that previously you couldn't.’ That just makes Canva AI so much more valuable, and the value immediately becomes clear. Since we're working in the space of creativity and productivity, having an organizational acceptance of the idea that ideas start in the chaos stage matters. You shouldn't judge ideas too harshly; instead, treat them as potentially promising candidates. And when you think about prioritization, it has to be thought about in proportion to how much you need to invest, and how much you need to continue to invest, to take it forward. That's a working model we've doubled down on.

How do pre-training and post-training the models work?

In terms of design, this is an area we've spent quite a few years on, with many iterations and improvements, so I'll just share how it looks right now. It's different from LLM training in some ways, but the core essence is still there. We're working with the Canva editor, breaking things down into objects.

So it usually starts with a lot of experiments, often with very small toy models and very small parameter counts. The benefit of that is we can learn and get results much, much more quickly. Obviously, they don't produce workable outputs.

You then do your pre-training run, and after the pre-training, you do things like reinforcement learning, supervised fine-tuning, and other fairly standard LLM techniques in the post-training stage. That part is not uncommon. But the difference is in the domain-specific nature and in having constraints and core focuses, because we can always rely on general-purpose LLMs for general-purpose tasks. We just need an LLM that's really, really good at the specialized thing.

How does Canva run its evals?

There are just a lot of things for an eval. It started with ELO-based ratings: out of these two presentations or Instagram posts, which ones did human raters prefer from a quick glance? I can't share all the complete details, as we obviously consider a lot of this confidential, but since then, we've evolved it to consider many facets, like medium awareness.

Where do you think creative AI is heading in the next two years?

If I were to dream a little bit further, I think there are a few ways this could play out. With the latest advances in AI, it's easy to imagine how visual communication and design become even more accessible to more people. Image generation already gives us a glimpse of that future, where creating something visual could start with a simple tool call. Like image generation makes it really easy to see how design might be a tool call.

What's really exciting is that as visual mediums become more accessible, the average person is able to show what they mean rather than trying to describe it. Visuals will become the new shorthand. Think about times when you previously wouldn't have thought of a visual artifact or visual communication as medium. But with AI, now that becomes possible, and you can share a prototype instead of a wall of text. This is what I'm really excited to unlock: the potential to turn so much of our everyday work—across nearly every industry—into visual communication that we wouldn’t even think to create today.

How is Canva using AI operationally beyond product and development?

We're pretty deeply ingrained when it comes to AI for everything. This also kind of helps with the feedback loop of actually building a better AI platform with design tools. One recent example is that we've started to adopt and build our internal finance workflows into Canva AI. These include everything from hooking together and adding connectors for the more enterprise systems that we already use for our financials, like our systems for tracking vendors, invoices, and payments.

It gives us very high-signal, high-quality feedback that has also cleared all of the really important training guardrails and consent, since we're all part of the same company. We use it to build Canva AI better for those use cases, and to start uplifting how all of our teams use it for operations, not just for building or shipping to our users. It's also about finding new use cases, and critically finding out what's actually useful, what actually saves time, what people say "this is something I love and will pay for," versus what's a bit of a distraction, a fancy demo that isn't what you actually asked for and doesn't really help you do your knowledge work.

How do you get the best out of yourself?

I'll start with the professional side, and something I've found really helpful and actually useful. When you're pushing boundaries and trying to do things, you often hit constraints. These could be constraints of capability, cost, or scale. Maybe your dream product experience is possible, but it's just not practical to scale to an audience as large as hundreds of millions of users like the one Canva serves every day.

I also use all the tools, so I'd say start playing around. I don't have a favorite, but making our core workflows effectively AI-accessible and giving it as much context as possible, past ideas included, has been super helpful.

On a personal level—and this is going to be a little bit of a curveball—honestly, one of my biggest personal developments in how I think has come from AI-free days. These are days when I don't use AI at all and try to put myself into a back-to-basics kind of view. To give you an idea, I was hacking around with a Raspberry Pi and some hobby embedded projects. In a setup where asking for assistance from a really capable AI system is so easy, you intrinsically take off a significant amount of cognitive load and burden, which is great. But it also doesn't challenge some of your important skills, as well as your talent and craft in many areas. So, having AI-free days is probably my top tip, as an AI guy, for both AI development and personal development.

Extra reading

• Vibe coding that matches your vision: Canva Code 2.0 is now available to all - July, 2026

• Introducing Canva AI 2.0: Reimagining how the world creates - April, 2026

• Canva’s Danny Wu on Giving 230 Million Users Superpowers With AI - July, 2025

And that’s it! You can follow Danny on LinkedIn or check out Canva on their website to keep up with what they’re building!

BRAIN FOOD 🧠 

TWEETS OF THE WEEK 🐣 

TOOLS WE RECOMMEND 🛠️

Every week, we highlight tools we like and those we actually use inside our business and give them an honest review. Today, we are highlighting Granola*—an AI-powered notepad that takes, summarizes, and organizes meeting notes without using intrusive recording bots.

See the full set of tools we use inside of Athyna & Open Source CEO here.

HOW I CAN HELP 🥳

P.S. Want to work together?

1. Hiring global talent: If you’re hiring tech, business or ops talent and want to do it 80% less, check out my startup, Athyna. 🌏

2. See my tech stack: Find our suite of tools & resources for both this newsletter and Athyna here. 🧰 

3. Reach an audience of tech leaders: Advertise with us if you want to get in front of founders, investors, and leaders in tech. 👀