Hi everyone 👋 Today, we’re excited to share more about how AI is integrated into Observable Canvases. We know, we know: you’re inundated with AI news every day. And skepticism around using AI for data analysis can be warranted. AI can hallucinate. It can get things wrong. When those mistakes happen in a black box, hidden from your view and impossible to verify, how can you trust AI outputs at all? At the same time, AI has the potential to transform how you work with data, from automating tedious data processing, to faster chart development, to revealing hidden patterns in complex data — if you can trust it. This is why we’ve taken a different approach. In canvases, AI is integrated, transparent, and verifiable so you can see exactly what’s happening with your data. We expose the code, reasoning, and intermediate steps behind each AI response. Because you can inspect AI’s outputs, you can more confidently use it to kickstart new data explorations, wrangle data faster with natural language prompts, and draft new charts that get added directly to your canvas. Plus, our AI understands the context of what’s already on the canvas, so it delivers outputs that are informed by, and most relevant to, your task at hand. With everything visible right on your canvas, you can see and make changes when AI gets something wrong. That lets you do fast, AI-powered data analysis and visualization — without a doubt. 🔆 Learn more in this new post from our founder and co-CEO Mike Bostock: https://observablehq.com/blog/playing-safely-with-fire-building-interpretable-ai-for-data-analysis