🍔 Let Claude Cook

I Built Something Faster Than I Could Buy the Domain

Santa Barbara Burger Week is back. Every year, dozens of restaurants across the city (and surrounding area, woo!) serve up $10 specialty burgers for a week long event, and the primary source of truth is a single article in The Independent.

I wanted a map and thought "I could build that!" In the past, this is where I'd buy a domain name, feel productive about it, make note to work on it in the evening, and never write a line of code. I have a graveyard of unused domains to prove it. But this time, I had a working app before the DNS finished propagating.

Check it out live at https://sbburgerweekmap.com

I Typically Don't Bother Actually Building These

Let me be honest: I would not have attempted this project last year, or heck even a few months ago. Not because it's technically hard, a map with some markers isn't rocket science. But the surface area of little tasks would have killed my motivation before I started:

• Scraping restaurant addresses from an article
• Geocoding every one of them into lat/lng coordinates
• extracting Google Maps and Apple Maps links
• Hand-wiring popup HTML for each location
• Getting the mobile layout right
• Setting up deployment

Any one of those is a 20-minute yak shave. Stack them all up and the project dies in the "maybe someday" pile, right next to the domain name I already bought for it.

Describe the What, Not the How

I described the outcome I wanted in plain English, and Claude Code figured out the implementation. I did not give pseudo-code, I didn't draw up architecture diagrams, I simply wrote out my intent.

My first prompt was basically: "Here's an article with a list of burger week restaurants. Build me an interactive map."

That's it. That was the "spec" doc.

id like to build a basic map application for Santa Barbara Burger Week. Ideally it would be a map with the locations and links to each location.
 
here is the source data i want to use to generate the webapp https://www.independent.com/2026/02/05/burger-week-2026/

I hit enter and went to go figure out a domain I liked. By the time I'd clicked purchase, I came back to a fully working map in my browser. Claude Code had:

• Fetched the article and extracted restaurants
• Geocoded every address using Nominatim
• Generated a data.js file with structured entries for each location
• Built a Leaflet map with color-coded markers by area
• Added a sidebar with search and filtering
• Made it mobile-responsive with a slide-up drawer

Just four files: index.html, app.js, style.css, and data.js. Open the HTML file and I've got a map. The domain purchase genuinely took longer than the first working version.

Iterating with Sentences

Once the base was working, I kept iterating with Claude running on a background terminal tab. This is where the workflow really clicked. I'd type a sentence, hit enter, and go do something else while Claude Code built it. Each feature was a quick prompt and then back to whatever I was doing:

"Make the map icon turn into a little 🍔 emoji when it's selected"

"When I hover over a restaurant in the list, show me where it is on the map"

"Have the area filter zoom the map to fit just those locations"

"Show a popup with burger details when I hover over a marker"

I wasn't writing code. I wasn't even watching it work most of the time. I was describing behavior, hitting enter, and coming back to a finished feature. Refresh the browser, check the result, type the next sentence. The implementation just appeared in the background, consistent with everything already built.

A few of my favorites:

• Cluster tooltips: Zoomed-out markers group into clusters showing a count. I asked for hover tooltips on clusters, and now they show a list of restaurant names. One sentence, done.
• Source monitoring: I asked for a GitHub Action that checks the Independent article daily for changes and opens an issue with a diff. It uses Python's stdlib HTML parser to extract just the article content (no pip dependencies), compares SHA-256 hashes, and commits the new baseline. I didn't spec any of that, I said "tell me when the article changes."
• Hand-picked map links: I wanted Apple Maps and Google Maps links that go directly to each business, not just an address search. Claude Code set up the data structure and fallbacks while I manually curated the links. We were literally working in parallel, I'd be on Apple Maps searching for a restaurant while Claude Code was building the next feature I'd asked for. By the time I'd pasted in a batch of links, there was a new feature waiting for me to review.

Pick Favorites (Because That's What You Do With the Print Copy)

I posted the project to reddit and got a comment telling me the first thing they do with the print article is cross off the restaurants they're not interested in. I found a new feature!

"Add a checklist mode where I can toggle restaurants on and off"

So back to the terminal, Claude Code added a "Pick Favorites" button that toggles checkboxes on the restaurant list. Unchecked restaurants dim to 40% opacity (both in the sidebar and on the map) so you can still see them but they fade into the background. Bulk "All" and "None" buttons for quick resets. Everything persists to localStorage, so your picks survive a page refresh.

The killer addition is Print Selected. Hit the button and it opens a new page with a numbered map and a grouped list of just your favorites - area by area, with addresses, phone numbers, and burger descriptions in a two-column layout. The map captures as a static image before printing so it actually renders in the PDF (Leaflet's CSS transforms don't survive print mode - a fun one to debug).

The whole feature took maybe three rounds of prompting. The print-to-PDF blank page bug took another two. Upholding the classic ratio: 20% building, 80% getting the edge cases right.

The Embed Got Picked Up

A local news site asked if they could embed the map in their burger week coverage. I didn't have an embed page, so:

"Build an embeddable version of the map for iframes"

Claude Code created a self-contained embed at /embed/map as its own HTML, JS, and CSS, but sharing data.js with the main site so restaurant updates propagate automatically. No header, no attribution footer, just the map with a compact sidebar. A showcase page at /embed has the iframe preview and copyable embed code.

Once the Pick Favorites feature landed on the main site, porting it to the embed was another one-liner. The embed shares the same localStorage key, so favorites picked on the main site carry over to the embedded version.

The Lack of Stack

The whole thing is:

• Leaflet + MarkerCluster (from CDN)
• CARTO basemap tiles
• Plain HTML, CSS, and JavaScript
• GitHub Pages for hosting
• A GitHub Action for monitoring the source article

It's really that simple, there is no React, no Tailwind, no build pipeline, just a single html file. This keeps something so simple, so very much, well, simple. I can iterate with Claude quickly and jump in to make manual edits on the fly.

That simplicity wasn't a constraint I ever explicitly set. Claude Code just reached for the simplest thing that works. When I wanted a burger emoji overlay on map markers, it didn't suggest swapping to a different mapping library. It created a divIcon overlay on top of the existing circleMarker.

So What Did I Do, Exactly?

It's worth being clear about what my contribution was, because this isn't a "the AI did everything" story:

• I had the idea. The article exists, the map doesn't. That's a human observation.
• I directed every feature. Each prompt was a product decision: what matters, what doesn't, what the user should see.
• I curated the data. I hand-picked Google Maps and Apple Maps links for 48 locations. I checked addresses. I verified coordinates. This was my main "job" and I did it while Claude Code built features in the background.
• I made taste decisions. Color palette, layout hierarchy, what goes in the popup, what goes in the sidebar.
• I said "not quite" a lot. The first pass was rarely the last. I'd see the result, adjust the direction, and iterate.
• I listened to a podcast. Seriously. I spent a good chunk of this project copying Apple Maps links while half-listening to a podcast and occasionally tabbing back to see what Claude Code had shipped while I wasn't looking. That's the vibe.

Rethinking My Capacity

The thing that surprised me most wasn't the speed. It was that I started at all.

This is a project I would have bookmarked, thought "that'd be cool," purchased the domain, and then never opened a code editor for. I know this because I've done it dozens of times. The activation energy was too high for what felt like a throwaway event.

My first prompt was barely a sentence. Thirty minutes later I had a live site. An hour later it had hover interactions, mobile support, source monitoring, and a README for anyone who wants to fork it for their own city's food event. Most of that hour, I was doing something else while Claude Code worked.

The barrier isn't knowing how to code anymore. It's knowing whether the thing is worth the effort to start. And when starting costs almost nothing, subscription aside, I can quickly prototype things I'd never build before.

The map is live at https://sbburgerweekmap.com. Fork it and map your own burger week!

Santa Barbara Burger Week 2026 runs February 19–25. All burgers are $10. No affiliation, just hungry.