Canva let candidates use AI and raised the interview's realism

Canva found a mismatch between its engineering work and its interviews. Almost half of its frontend and backend engineers were daily users of AI coding tools, yet the old Computer Science Fundamentals interview asked candidates to write algorithms unaided. Canva's own experiments found that AI could produce correct, documented solutions to traditional questions in seconds, while covert AI use was increasingly difficult to police.

The company chose transparency over detection. For backend, frontend, and machine-learning roles, it piloted an AI-Assisted Coding interview in which candidates were expected to use their preferred tools. The questions became more complex, ambiguous, and realistic—for example, designing a control system for aircraft at a busy airport rather than implementing Conway's Game of Life—so a single prompt could not settle the problem.

The interview evaluated whether candidates clarified requirements, used AI strategically, maintained control of the overall design, found defects in generated code, and held the result to production standards. Canva reported that its strongest candidates treated AI as a multiplier rather than an answer machine; candidates with little AI experience often struggled with the judgment needed to guide or challenge it. The pilot felt more engaging and, in Canva's assessment, produced stronger predictive signals about performance.

Test judgment with AI, not memory without it

Banning tools that engineers use every day makes the interview measure isolation from the job rather than performance on it.

Ambiguous product problems force candidates to clarify requirements and make trade-offs that a memorized algorithm question hides.

Requiring candidates to own the generated code preserves technical rigor while making AI collaboration observable.

Reviewing and debugging generated code tests the comprehension and judgment engineers need after AI writes the first draft.

Transparency removes the cat-and-mouse game of detection and turns tool use into an explicit competency.

When a tool changes the job, do not defend an obsolete test by banning the tool; redesign the task so the scarce judgment remains visible and accountable.

Canva told candidates ahead of time that AI use was expected and encouraged them to practise with the tools. The company framed the redesign as part of an AI-everywhere operating philosophy, while keeping code fluency, technical depth, and ownership as non-negotiable requirements; it also committed to revisiting the format as model capabilities change.