LLenny's Podcast
← All frameworks
InnovationCameron Adams (co-founder and CPO)

The Three-Pillar AI Integration Model

Build where you have data advantage, partner for commodities, and open an app ecosystem for the rest

Difficulty
Advanced
Time to result
~months to results
Steps
4
Confidence
88%

Canva integrates AI through three deliberate pillars rather than building everything or wrapping a single model. Build in-house only where you have the biggest data and insight advantage; partner with the world's best providers for commoditised capabilities like LLMs; and let an app ecosystem bring the long tail of AI features to your users. AI must serve what users are trying to do, not be a chatbot slapped on top.

Origin

Cameron Adams describes Canva's approach after seven years of having machine-learning engineers, framing AI as 'the next way of democratizing design.'

Core principles

  • 01AI can't be the whole product — don't just be a wrapper around an LLM or slap a chatbot on existing features
  • 02Start from what people want to do and how AI helps them reach the goal faster, then choose the pillar
  • 03Build only where you have the biggest data, insight, and criticality advantage
  • 04LLMs are a commodity — partner rather than reinvent what others do far better with more resources
  • 05Your existing audience is a magnet for third-party developers who extend your product

How to run it

  1. 1

    Build in-house where you hold the data advantage

    Develop your own AI tech only for areas where you have the most data, the most insight, and the most criticality to your product — for Canva, models around design and images.

  2. 2

    Partner for commodity capabilities

    For capabilities that are now commodities (e.g. LLMs), find the world's best providers and integrate them well rather than internalising them — Canva partnered with OpenAI and with Runway ML for video generation.

    Watch out Building your own LLM wastes resources on something providers do better and cheaper — treat it as a commodity.

  3. 3

    Open an app ecosystem for the long tail

    Let third-party developers build apps that integrate directly with your product, giving them access to your large audience while extending your AI surface (music generators, virtual avatars, and more).

    Pro tip A large monthly audience (Canva had 170M) makes developers eager to build integrations for you.

  4. 4

    Keep AI subordinate to user goals

    Across all three pillars, keep the focus on helping people reach their big goals naturally, so AI feels like part of the experience rather than technology pushed in their face.

    Watch out Deeply think about how AI gets users to the goal faster — don't just bolt a chatbot onto what already existed.

In the wild

Canva's three pillars in practice

Canva builds its own models for design and images (data advantage), partners with OpenAI and Runway ML for LLM and video generation (commodities), and runs an app developer ecosystem where AI developers ship music generators and presentation avatars to Canva's 170M monthly users.

A coherent AI experience that keeps focus on user goals and extends Canva's mission of democratizing design.

Common mistakes

Being a thin wrapper around an LLM

A product built purely on AI or as an LLM wrapper skips the hard work of meeting what users actually need.

Building commodity AI yourself

Reinventing an LLM burns resources on something specialist providers do far better; partner instead and reserve build effort for your data-advantaged areas.

Is it for you?

Best for

Established product companies with a proprietary data advantage and a sizeable user base deciding how to add AI

Not ideal for

Early-stage teams with no distinctive data moat or audience, where the build-vs-partner-vs-ecosystem split doesn't yet apply

From the transcript

it also can't just be the basis for your product you can't just be a product that's purely built on AI on being being a…

55:00

we focus on building the AI Tech that we have the biggest advantage in that we have the most data that we can put into…

56:30

you don't need to create an l because it's a commodity thing now and there's a bunch of providers who can do it way better

57:00

finding a great partner like open AI we partnered with Runway ml to do video generation

57:30

for us the third pillar is our app ecosystem so we're fortunate now we've got 170 million people using the product every single month

57:30

From the episode

Inside Canva: Coaches not managers, giving away your Legos, and running profitably

Cameron Adams (co-founder and CPO)