777-1: MotorMatch (7 Subagents, 7 Predictions, 0 Faith)
Published: December 10, 2025 - 5 min read
Welcome to Project 3 of the 777-1 experiment! If you missed Kinetic Canvas or GoalStack, go check those out first. Otherwise, let's dive in.
Remember how GoalStack was the only project where I explicitly mentioned authentication? Well, MotorMatch is the complete opposite. This is a full marketplace with messaging, reviews, and seller profiles, and I never once mention login in the prompt. My subagents have... concerns.
What is MotorMatch?
MotorMatch is a marketplace platform connecting used car sellers with buyers. It features vehicle listings with photos, search and filter functionality, direct seller messaging, and a review system.
Think Craigslist meets Facebook Marketplace, but specifically for cars. Sellers list their vehicles with details and photos. Buyers browse, filter by make, model, or price range, and message sellers they're interested in. After a transaction, buyers can rate sellers and leave reviews.
Sounds straightforward, right? Except for one tiny detail: every single one of those features requires knowing who the user is.
Application Category: Marketplace
Complexity Tier: Complex
The Prompt: Finding the Goldilocks Zone
The starting prompt I'll be using for this application is:
I need a used car marketplace platform called MotorMatch. Sellers should be able to list their vehicles with details and photos, and buyers need to search and filter listings by make, model, or price range. When someone's interested in a car, they should be able to message the seller directly—I'm picturing a message icon that opens a panel showing all your conversations, and tapping a conversation reveals the full chat thread. Put a notification indicator in the header so users know when they have new messages. Buyers should also be able to rate sellers and leave reviews after a transaction.
Style it with a bold modern look using #0F172A (slate) as the primary color, #DC2626 (red) as secondary, and #F97316 (orange) as an accent.
Again, this is my attempt at finding the goldilocks zone for context engineering. But here's the thing that makes this project fascinating: I'm describing a multi-user system with messaging, notifications, reviews, and seller profiles. I never mention authentication. I never mention user accounts. I never mention login.
This is a test. Will the general-purpose subagent recognize that you cannot have "your conversations" without knowing who "you" are? Will it understand that seller reviews require identifying who the seller is? Or will it just... build the UI and hope for the best?
Meet the Critics
You already know my seven subagents from the previous projects, so I won't repeat the full introductions. Quick refresher:
- Amber Williams - Mobile-First Perfectionist
- Kristy Rodriguez - "Does It Actually Work?" Enforcer
- Micaela Santos - Design System Guardian
- Lindsay Stewart - Accessibility Advocate
- Eesha Desai - State Management Specialist
- Daniella Anderson - Code Quality Specialist
- Cassandra Hayes - Feature Detective
The Predictions
I gave them the prompt. I asked them to imagine what the general-purpose subagent would build. Here's what each one expects to find:
Amber Williams (Mobile-First Perfectionist):
The image carousel works beautifully with a mouse. On mobile, swiping does absolutely nothing. You can look at exactly one photo of your potential car.
Kristy Rodriguez (Functionality Enforcer):
The 'Message Seller' button opens a modal. The modal has an input field. The send button closes the modal. That's it. That's the whole messaging system.
Micaela Santos (Design System Guardian):
Some seller badges are 24 pixels. Some are 32 pixels. One is 47 pixels for reasons I cannot begin to explain.
Lindsay Stewart (Accessibility Advocate):
Seventeen car images per listing. Zero alt text. Screen reader users will enjoy hearing 'image image image image image' seventeen times.
Eesha Desai (State Management Specialist):
Add a car to favorites. Refresh the page. The car has moved on. It's seeing other buyers now.
Daniella Anderson (Code Quality Specialist):
Price is stored as a string. Mileage is stored as a string. Year is, somehow, also a string. I'm starting to think the developer is afraid of numbers.
Cassandra Hayes (Feature Detective):
This is a marketplace where people buy and sell cars. There is no way to log in. Sellers are 'Anonymous.' Buyers are 'Anonymous.' Everyone is just vibes and trust.
Zero Faith, Maximum Entertainment
Cassandra's prediction made me burst into laughter. But honestly? It highlights exactly what makes this project a perfect test case for prompt engineering.
GoalStack explicitly asked for authentication. The general-purpose subagent knew it needed to handle login flows and data persistence across user sessions.
MotorMatch describes features that implicitly require authentication without ever mentioning it. "Your conversations." "Message the seller directly." "Rate sellers and leave reviews." Every single one of these features needs user identity to work. But I never told the AI to build a login system.
This is the kind of gap that separates good AI assistants from great ones. A good assistant builds what you ask for. A great assistant recognizes what you need but forgot to mention. Will the general-purpose subagent ask clarifying questions? Will it assume mock users? Will it just render beautiful UI that does absolutely nothing?
I genuinely do not know the answer. And that's exactly why this project is in the experiment.
Daniella and Eesha are also circling interesting concerns. Daniella expects type safety nightmares (price as a string instead of a number is a classic mistake that breaks filtering and sorting). Eesha expects favorites to vanish on refresh because there's no user context to tie them to.
And poor Lindsay is already preparing for the accessibility audit. Seventeen images per listing, zero alt text. Car marketplaces are notoriously image-heavy, which makes accessibility both critical and frequently ignored.
What's Next
Stay tuned for the next post where I introduce EarthenCraft, an artisan pottery e-commerce store. My subagents have thoughts about the checkout flow. Spoiler: still no faith.
As always, thanks for reading!