Build a Researched Isochrone Map with Claude 5 Fable

What this is, in plain English

An isochrone map shows how far you can travel from a place in a given time: lines connect all the points you can reach in, say, two hours, four hours, and so on. This workflow is a real example of handing one big, well-described task to a top-tier AI (Claude’s Fable model, one of its most advanced) inside Claude Code, and letting it research the data and build the whole interactive map over several hours, mostly on its own.

Be honest about what this is: it is more a demonstration of what a frontier AI can do than a step-by-step recipe with a guaranteed result. It needs access to a very capable model (the author had early access), it runs for hours, and the output is a one-off creative project that will differ each time. The reusable lesson, which anyone can apply, is the technique: how to brief a powerful AI on an ambitious build and steer it with light feedback.

What you can use it for

The “give a frontier model one ambitious, well-specified project and let it run” approach suits any rich, research-heavy build:

• Data visualizations. Interactive maps, charts, or dashboards that need both real research and good design.
• Self-contained mini-apps. A calculator, explorer, or tool built from a single detailed brief.
• Research-then-build tasks. Anything where the AI must gather many real facts first, then turn them into something usable.
• Design exploration. Ask for a distinctive visual style (the author’s map borrowed the look of an 1881 isochrone map) and let the model run with it.
• Seeing the ceiling. A way to feel what the most capable models can do before you commit to a bigger project.

Tools you need

• Claude (freemium): you work inside Claude Code (the coding environment) and, ideally, use the most advanced model you can access. The author used the Fable model in early access; the free and standard tiers can still attempt this, with less depth.

How it actually works

The whole thing is really one big prompt plus gentle steering. The realistic shape:

1. Open Claude Code with the strongest model you have. Choose the most capable Claude model available to you. Results scale with the model.

2. Give it the full brief in one prompt. Be detailed about data, travel modes, and design:

   I want you to build a fully researched and beautiful isochrone map that lets me pick various cities and see real isochrone lines based on real data. I want the design to be unique. Take into account airports (and travel time to and from airports), trains, walking, and driving. The data does not need to be live but should be real, based on your research. You can start with a few cities but more general is better. This should be an entirely new project.

3. Agree on a style and let it run. The model may propose a design direction (the author’s suggested the 1881 isochrone style). Approve it and let it work through its multi-hour build.

4. Steer with light feedback. A nudge or two is usually enough, for example:

   Make it better by refining the visual clarity of the travel lines.

5. Review the result. Examine the finished map, check the travel times look sensible, and ask for specific fixes where needed.

Words you’ll see, explained

• Isochrone map: a map whose lines join all the places reachable within the same travel time.
• Claude Code: Claude’s coding environment, where the model can build and run a project, not just chat.
• Frontier model: one of the most advanced AI models available (the Fable line here); these handle long, complex jobs better than smaller ones.
• Iterate: improve something in rounds, giving feedback and letting the AI refine, rather than expecting it perfect first try.

Original source

Based on Ethan Mollick’s post “What it feels like to work with Mythos”, sharing his early-access experience using Claude’s Fable model to build a researched isochrone map.

Notes & variations

• Do you even need the top model? The depth of research and polish here came from a frontier model. On a free or standard tier you can still build a simpler version; just expect less sophistication.
• Common mistake: expecting a finished result from a single prompt. Even here it took a detailed brief plus a couple of follow-ups. Plan to iterate.
• Tip for better results: front-load detail (cities, travel modes, data sources, the look you want), then steer with short, specific feedback rather than vague “make it nicer”.