Generating running routes with GPT-6 Astra and ChatGPT Work
Simon Willison’s Weblog Simon Willison
ChatGPT Work used OSM data to build 5K and 10K loop routes from one address. It took 27 minutes and spit out maps plus GPX and GeoJSON files.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Simon Willison says he had ChatGPT Work, running GPT-6 Astra (Max), build running routes from his home using OpenStreetMap data. He asked for 5K and 10K loops that started and ended at his address, and the system got there after 27 minutes.
What makes the result interesting isn’t the prompt, but the shape of the output. It didn’t just answer in text. It produced an embedded visualization, plus downloadable GPX and GeoJSON files. That’s the kind of package that feels less like a chatbot reply and more like a small planning tool someone would actually use.
Willison framed it as a neat thing he’d tried that morning, which sounds about right. The request is specific, the data source is public, and the result is practical. There’s no grand manifesto here, just a reminder that these systems are increasingly good at taking a messy real-world task and turning it into something structured enough to use.
And because the task is about local routes from a precise address, the demo also shows how quickly these models are moving beyond generic text generation. They can now stitch together data, format it for different uses, and hand back files a human can load elsewhere. That’s quietly useful, which is usually where the real value hides.
My take — AI-written commentary, not fact-checked reporting
This is the sort of AI demo that actually matters: not another fluent paragraph, but a useful file. The hype crowd loves grand claims; the rest of us should be watching for tools that save 27 minutes and don’t require a ceremony. Open models can get there too, but only if people stop treating “chat” as the whole product.
Read more about this at: Simon Willison’s Weblog
