Overview
Boston Last Call is a web app that helps restaurants, cafes, bakeries and grocery stores in Boston sell leftover food near closing time instead of throwing it away. Stores post discounted items at the end of the day. Nearby customers reserve them and pick them up during the store’s pickup window, paying at the counter.
I am building the entire project in collaboration with Claude Cowork, an AI agent that helped me research places, write code, and style the site. It is my first time building a full product with an AI as a partner, and it has completely changed how I think about what one person can create.
**Keep in Mind: This case study and project is a Work in Progress!**
The Goal: Less Waste, Cheaper Food
Last Call has two sides, and I designed it for both:
For customers: Browse today’s deals on a list and a map, see how much you save, reserve one to five items, and get a five-letter pickup code. Changed your mind? Cancel, and the items go back on sale for someone else.
For store owners: Sign up as an owner, set up your store (name, type, address, a pin on the map, and your usual pickup window), then post items with a quantity, a regular price, a deal price and a pickup day up to a week ahead. Owners see each reservation with its pickup code and mark it “Picked up” or “No-show”. A no-show puts the items back on sale.
Working With AI as a Teammate
The biggest change in this project was learning to treat AI as a collaborator rather than a tool. I wrote a detailed project guide (a CLAUDE.md file) that works like a creative brief and rulebook for the agent. It describes how the app works, how the data is stored, the writing style, and hard rules, like never inventing facts. The agent also keeps a “Current Status” section updated at the end of every session, so each new session starts where the last one stopped.
To test with real-looking data, I picked 16 real Boston stores in Mission Hill, Back Bay, the Seaport, the North End, Chinatown, East Boston and Cambridge, and used them as demo listings. Each one clearly says the store hasn’t joined the app. I never wanted to suggest a real business had signed up when it hadn’t.
This setup taught me that the quality of AI output depends heavily on the quality of the direction you give it. The clearer my instructions were, the better and more consistent the results became.
Designing the Experience
Design is still my favorite part. I gave the app the look of a grocery app: a green palette with category chips, a hero banner, rounded product cards with discount badges, and food photos or emoji tiles when an item has no photo. I kept a deep brick red for buttons so actions stand out, and designed a logo (a shopping bag with a clock and a leaf) that also works as the favicon.
Working with AI made this feel like iterating in Figma. I could share references, describe changes in plain language, and see them almost right away. Some small details made a big difference: countdowns showing how long until pickup, map dots that are colored when a store has deals and grey when it doesn’t, and a reserve dialog with the photo on top and the details in clear labeled rows.
Where AI Fell Short
AI sped things up a lot, but it wasn’t perfect!
Real places, but no real photos. Claude was very accurate at finding real Boston stores, with correct addresses and locations. Photos were a different story. It couldn’t find good images of the stores’ actual food, so it filled in generic placeholder illustrations and emoji instead. I’ve been finding real photos myself and adding them one store at a time.
Blocked websites. Some websites block AI agents, so Claude couldn’t always open a store’s site to check details. When that happened, I had to look things up myself.
Getting the design exactly right. Claude could get close to the look I described quickly, but not always to the exact design in my head. I often had to follow up with small fixes like “add more space between the cards” or “make this green a little darker.” The final design came from lots of small rounds of feedback, not one perfect prompt.
AI is a fast and capable, but it still needs a human to check its work and keep the vision on track.
Ideas I’m exploring next:
Favorite stores, so customers hear about deals from places they love
Pickup reminders before a pickup window closes
A weekly history for owners, showing what sold, what didn’t, and how much food was saved (dashboard possibly!)
What I learned
This project is showing me how far AI has come and how much faster it lets me build. Things that would normally take me days, like setting up a working map, building accounts with two roles, or styling a whole interface, now take hours. Instead of being limited by what I know how to code, I’m limited only by what I can imagine and clearly describe.
It also showed me that AI doesn’t replace the developer, it changes the job. My work became defining the vision, deciding what the product should and shouldn’t be, setting rules, and checking the work. It feels like being a product manager, a designer and an economist at once: thinking about pricing, incentives, and what makes both a store owner and a student want to use the same app.