Artificial intelligence is radically transforming the landscape of software development, and the most innovative companies are already adopting ambitious strategies to integrate these technologies. At Canva, a graphic design platform used by millions of users, this transformation takes on a particularly impressive dimension. Brendan Humphreys, the company’s CTO, reveals striking figures: 80% of Canva’s engineers report being more productive thanks to AI tools. This success is not a matter of chance, but rather a well-thought-out strategic approach that combines freedom of experimentation, rigorous security assessment, and personalized support for teams. With 2,300 engineers using tools like Sourcegraph, Cursor, Claude, or ChatGPT on a daily basis, Canva demonstrates that massive adoption of AI is not only possible but also profitable. This unique experience offers valuable insights for all companies looking to leverage the revolution of artificial intelligence in their development processes.
đ Summary
đ In brief
- 80% of engineers at Canva report improved productivity thanks to AI
- 2,300 engineers use tools like Cursor, Claude, and ChatGPT daily
- Liberal adoption policy with permissive licenses and pay-per-use pricing
- Rigorous security assessment before large-scale deployment
- Differentiated usage based on developers’ experience levels
A liberal and strategic adoption policy
At Canva, the approach to artificial intelligence is distinguished by its philosophy of openness and experimentation. Brendan Humphreys explains that the company has deliberately chosen a very liberal policy regarding AI adoption, encouraging every employee to experiment with these tools to understand how to apply them to their daily work. This strategy is based on the belief that innovation arises from the freedom of exploration and the creativity of teams.
The company provides its 2,300 engineers with an impressive range of AI tools: Sourcegraph for code analysis, Cursor for development assistance, Claude and ChatGPT for code generation, as well as many other specialized tools. This diversity allows each developer to find the solutions best suited to their specific needs and work style.
A particularly innovative aspect of this approach concerns the adopted business model. Rather than paying for unused licenses, Canva favors a consumption-based pricing model. This strategy not only optimizes costs but also encourages more thoughtful and efficient use of tools. Engineers can easily access the resources they need without excessive budget constraints.
The company also maintains direct relationships with key AI tool providers, allowing it to anticipate technological developments and quickly adapt its strategy. This active technological monitoring, combined with a community of engaged engineers and a responsive IT department, creates an ecosystem conducive to continuous innovation.
Security and evaluation of AI tools
Despite this liberal approach, Canva has not neglected the crucial security aspects related to the adoption of artificial intelligence. The company has developed a sophisticated supplier analysis and testing program, ensuring that each integrated tool meets the highest security standards. This rigorous process is an essential prerequisite for any large-scale deployment.
The evaluation process follows a structured methodology in several steps. First, the security team conducts a thorough analysis of data movements around AI models. This step allows for a precise understanding of what information is transmitted, how it is processed, and whether it is used for training the models. This transparency is essential for protecting the company’s intellectual property and the sensitive data of its clients.
Threat model evaluation represents another fundamental aspect of this process. Security teams analyze the potential risks introduced by each LLM, from technical vulnerabilities to data leakage risks. This proactive approach allows for the identification and mitigation of risks before they become problematic. I find this approach particularly exemplary, as it demonstrates that innovation and security can coexist perfectly.
Once the security evaluation is validated, the company launches a limited pilot project to test the tool in real-world conditions. This phase allows for quick feedback on the tool’s suitability for the company’s specific needs while minimizing risks. Only after the success of this pilot phase does Canva consider a broader deployment with permissive licenses for the entire technology department. This gradual approach ensures a successful and secure adoption of new artificial intelligence technologies.
Productivity metrics and return on investment
One of the most remarkable aspects of Canva’s experience lies in its ability to concretely measure the impact of AI on productivity. Unlike many companies that settle for superficial metrics, Canva has developed a more nuanced and realistic approach to performance evaluation. Brendan Humphreys expresses his scepticism towards overly simplistic statistics, such as the percentage of code generated by AI, which he considers to be unrepresentative of reality.
The company prefers to evaluate engineers’ subjective perceptions of their own productivity. This approach, while less quantifiable, offers a more authentic view of the real impact of AI tools on daily work. The result is telling: 80% of engineers report feeling more productive when using these tools. This metric, combined with the observation of an increase in workload among AI users, provides a reliable indication of the effectiveness of these technologies.
A particularly interesting benefit identified by Canva concerns the improvement of developers’ flow state. AI tools act as a constant pair programmer, allowing engineers to maintain their focus for longer periods. Previously, an interruption to fix a bug or search for information could easily lead to juggling multiple tasks simultaneously, breaking concentration. With AI, developers can delegate certain tasks while remaining focused on their primary goal.
This improvement in concentration translates into higher quality work and increased job satisfaction. Engineers can spend more time on the creative and strategic aspects of their work, while AI takes care of more repetitive or technical tasks. This division of roles not only optimizes productivity but also contributes to the professional fulfillment of teams. Canva’s approach demonstrates that AI tools can revolutionize work methods far beyond mere automation.
Differences in usage between junior and senior developers
Analyzing AI usage patterns at Canva reveals fascinating differences between junior and senior developers. These variations in approach perfectly illustrate how professional experience influences the way individuals integrate new technologies into their workflow. This understanding of differentiated usage allows the company to adapt its training and support strategies.
Junior engineers primarily use AI for prototyping and identifying technical solutions to their problems. This approach reflects their need for learning and their desire to quickly understand best practices. AI serves as their virtual mentor, guiding them in discovering new approaches and helping them overcome technical obstacles. This educational use of AI significantly accelerates their skill development and allows them to contribute more quickly to complex projects.
Experienced developers, from intermediate to senior levels, adopt a different approach, using AI as a true superpower integrated into their development environment. They leverage these tools as a sophisticated pair programmer, capable of reviewing their work and generating code with advanced contextual intelligence. This more mature usage capitalizes on their existing expertise to maximize the efficiency of AI tools.
This differentiation in usage raises important questions about the future of developer training. I believe that Canva’s observation demonstrates the importance of maintaining a balance between technological assistance and the development of fundamental skills. Junior developers must continue to acquire a deep understanding of programming principles while learning to leverage AI tools. This hybrid approach ensures comprehensive training and prepares future engineers to thrive in a constantly evolving technological environment. Canva’s experience suggests that AI amplifies existing skills rather than replacing them, creating new opportunities for all experience levels.
The future of AI in software development
Canva’s vision for the future of artificial intelligence in software development is characterized by a pragmatic realism that contrasts with the sometimes utopian discourse in the industry. Brendan Humphreys adopts a measured position, recognizing both the immense potential of these technologies and their current limitations. This balanced approach offers valuable insight into the future developments of AI in the software industry.
The company is currently experimenting with more autonomous code agents but is facing significant technical challenges. With tens of millions of lines of code, Canva’s codebase exceeds the capabilities of most current commercial tools. This reality has pushed the company to develop internal solutions and collaborate closely with providers to improve their products. This collaborative approach illustrates the importance of strategic partnerships in the evolution of AI.
Regarding complete automation, Canva maintains a clear position: these tools must remain under human supervision. The company recognizes that AI models can hallucinate and make mistakes with disconcerting confidence, requiring constant human validation. This cautious approach prioritizes the use of AI to generate solution proposals that are then reviewed, tested, and validated by experienced engineers.
The question of replacing developers, particularly juniors, with AI is a topic of deep reflection at Canva. The company categorically rejects this possibility in the medium term, emphasizing that coding represents only a tiny part of the software engineer’s job. Creative aspects, aesthetic judgment, quality assessment, and accountability for the final result remain fundamentally human. This humanistic vision of technology positions AI as a tool for augmentation rather than substitution, preserving the unique value of human expertise in the software development process. Canva’s approach to intelligent automation offers a balanced model for the industry.
Conclusion
Canva’s experience with artificial intelligence offers a remarkable model of successful integration of these technologies in a large-scale software development environment. The results achieved, with 80% of engineers reporting improved productivity, demonstrate that a strategic and well-executed approach can generate tangible benefits. The combination of a liberal adoption policy, rigorous security assessment, and personalized support for teams constitutes a reproducible framework for other organizations.
I am particularly impressed by the maturity of Canva’s approach, which avoids the pitfalls of hype to focus on meaningful metrics and concrete benefits. Their recognition of the current limitations of AI, combined with an optimistic yet realistic vision of its future potential, offers a balanced perspective that is often lacking in discussions on this topic. This experience demonstrates that AI can indeed transform đ work methods without replacing human expertise, rather creating a productive synergy between artificial intelligence and human intelligence. The future of software development looks promising, provided that a thoughtful and responsible approach to these revolutionary technologies is adopted.
