SUMM

Excalidraw was created during a period of active procrastination, originally intended as a tool for drawing illustrations for blog posts.
The inspiration came when an existing tool (Zibl) was offline, prompting the development of a rough prototype using Rough.js.
Sharing the prototype on Twitter led to unexpectedly high interest and engagement, leading to the ongoing development of Excalidraw.
The shift to remote work during the COVID-19 pandemic led people to use Excalidraw for online whiteboarding, a use case the creator had not initially considered.

Existing online whiteboard tools initially mimicked the physical limitations of actual whiteboards, such as fixed size and pixel-by-pixel erasing.
Excalidraw broke this pattern by allowing infinite canvas scrolling, simple object selection/deletion, text boxes, and easy shape creation.
Features like copy-pasting, which are impossible on physical whiteboards, became essential in the virtual context.
This experience highlighted the importance of rethinking digital tools instead of simply translating physical metaphors.

Many current AI integrations in products simply mirror existing workflows, often leading to poor user experiences.
There is a need to explore unique, AI-native interactions instead of just layering AI onto old paradigms.
As the field is new, sharing lessons at conferences helps accelerate the evolution of AI-driven products.

Initial attempts to integrate AI-generated realistic images into Excalidraw weren't useful, as users mainly needed diagrammatic, not realistic, drawings.
Lesson: Not every available AI feature is appropriate for every product; indiscriminate integration can negatively impact the user experience.
Attempts to use language models to directly generate diagram files failed due to inaccurate outputs.
However, generating domain-specific languages for diagrams (e.g., Mermaid JS syntax) using LLMs proved effective, allowing structured diagrams to be created from text prompts.

By translating Mermaid JS outputs into Excalidraw diagrams, users could prompt creation of complex diagrams (flowcharts, org charts, architecture diagrams) and then refine them manually.
This integration exemplifies a productive human–AI partnership: AI generates a strong starting point, and the user personalizes the result.
Anticipation of future features: more interactive back-and-forth between user edits and further AI refinements.

Automatic document naming: LLMs suggest meaningful file names based on document content, reducing user effort.
Upcoming illustration generation: enables users to generate infographic-style illustrations to enhance their diagrams and presentations.
Background removal in logos: identifying this as a pain point, the speaker invites the community to develop a robust, browser-based background removal tool for images.

As AI-native product design matures, teams should innovate on interaction models instead of mimicking past paradigms.
Developers should resist the urge to include every AI model; only features that add real user value should be shipped.
LLMs perform best when provided with structured, domain-specific languages, facilitating accurate and useful outputs for downstream applications.

Excalidraw: AI and Human Whiteboarding Partnership - Christopher Chedeau