Canva’s AI assistant can now call various tools to make designs for you

The latest iteration of Canva’s AI assistant is now capable of invoking external tools to generate complex designs. This functionality moves the assistant beyond simple text generation into dynamic task execution. The core change is the integration of functional API calls into the prompt-to-design workflow.

This capability shifts the focus from static image generation to dynamic agentic workflows. When running agentic tasks, the success metric changes from simple token quality to successful multi-step tool invocation. For instance, managing the inference cost for a multi-step RAG workflow involving multiple external APIs requires careful monitoring of latency, potentially pushing costs past $100 per complex task, as seen in some internal testing. Developers often assume tool calling is cheap and instantaneous, but introducing external latency into a production pipeline demands dedicated infrastructure monitoring.

Teams running multi-step design generation agents must establish strict latency budgets for external tool calls before deploying. Only teams focusing on optimizing inference time for specialized tools like Haiku instead of GPT-4 can maintain production velocity. Infrastructure teams must now focus on API gateway costs rather than just model weights. Ignore this if your system requires high-throughput, multi-tool execution.