El Bruno

Posted on Jul 28 • Originally published at elbruno.com on Jul 7

AI Agents Built My Shopping Cart (And I Just Watched) 🤖

#englishpost #artificialintelligen #blazor #codesample

Hi!

FYI: This was written with some AI help 😂

Okay, so picture this: I’m sitting here with my morning coffee, and I decide to add a shopping cart feature to the eShopLite demo scenarios. But instead of cracking my knuckles and diving into code like usual… I just watched AI agents do ALL the work. 🤯

I’m not kidding. In 8 minutes, I recorded the entire journey from “hey, we need a shopping cart” to a complete, production-ready e-commerce feature. And honestly? It blew my mind.

TL:DR: here is the video

The Setup: eShopLite Needs a Shopping Cart

So I was playing around with eShopLite – our sample e-commerce app that’s built with .NET Aspire, Blazor, and has this cool Semantic search feature using Azure AI Foundry models.

Pretty neat stuff, but it was missing something obvious: you couldn’t actually buy anything. 🛒

The old me would’ve fired up Visual Studio, and start coding. But 2025 me? I decided to see what happens when you let AI agents handle the entire thing.

Round 1: Claude Sonnet Gets the Requirements Right

First up was Claude Sonnet 4 in VS Code. I basically said “Hey, look at this codebase and write me a proper Product Requirements Document for a shopping cart feature.”

What came back was… honestly incredible. This wasn’t some half-baked bullet point list. We’re talking about a full 12-section PRD with:

📊 Executive summary with actual success metrics
⚡ Functional requirements that covered everything from cart management to checkout flows
🏗 Technical architecture with real code examples (not pseudo-code!)
📅 4-phase implementation plan broken down by weeks
🧪 Testing strategy covering unit tests, integration tests, the works
⚠ Risk assessment with actual mitigation strategies

Check out some of the entity models it generated:


public class CartItem
{
    public int ProductId { get; set; }
    public Product Product { get; set; }
    public int Quantity { get; set; }
    public decimal UnitPrice { get; set; }
    public decimal TotalPrice => Quantity * UnitPrice;
}

public class Cart
{
    public string SessionId { get; set; }
    public List<CartItem> Items { get; set; }
    public DateTime CreatedAt { get; set; }
    public DateTime UpdatedAt { get; set; }
    public decimal TotalAmount => Items.Sum(i => i.TotalPrice);
    public int TotalItems => Items.Sum(i => i.Quantity);
}

Round 2: GitHub Copilot Becomes Project Manager

Next, I told GitHub Copilot to take all that PRD goodness and create a proper GitHub issue in the Azure-Samples/eShopLite repo.

And wow, it didn’t just slap together some basic “add shopping cart pls” issue. It created Issue #31 with:

🎯 Complete technical specifications
📋 Implementation phases with clear deliverables
🏷 Proper labels (enhancement, feature-request, shopping-cart, e-commerce)
✅ Detailed acceptance criteria
📚 Links to all the relevant docs

Round 3: GitHub Copilot Coding Agent Goes Full Beast Mode

Instead of coding it myself, I assigned the issue to GitHub Copilot Coding Agent.

I literally just typed:

Add the Issue #31 to @copilot”

Then I sat back and watched the magic happen. This AI agent:

🔍 Analyzed the entire codebase to understand how everything worked
🌿 Created a new branch (copilot/fix-31) like a good developer
🏗 Built multiple projects – CartEntities, service interfaces, Blazor components
💻 Implemented EVERYTHING with proper error handling and validation
📝 Created a pull request with detailed docs explaining all the changes

Plot Twist: I Became the AI Whisperer 🎭

Now, don’t get me wrong – I wasn’t just sitting there doing nothing. My role shifted to something way more interesting, because, in every iteration, once GH Copilot Coding Agent generated some code, I started to do this:

🎯 Requirements : Making sure the AI “got” what we actually needed.
👀 Code Reviewer : Spotting edge cases and suggesting improvements, and hey, marked down some errors!
🎨 UX Advocate : Ensuring the user experience didn’t suck
🛡 Quality Guardian : Making sure this was actually production-ready

The conversation with the Copilot agent was surprisingly natural. When I said “hey, the checkout validation could be better,” it immediately understood and enhanced the implementation. When I mentioned mobile responsiveness, boom – updated CSS on the spot.

It felt less like programming and more like directing a really smart intern who happens to code at superhuman speed. 🚀

What Actually Got Built

The final result? A complete shopping cart system that would make any e-commerce developer proud:

🛠 Tech Stack

.NET + Aspire : Microsoft’s cloud-native platform
Blazor Server : For those sweet real-time UI updates
Session Storage : Using ProtectedSessionStorage (secure guest checkout)
Bootstrap + Font Awesome : Because we’re not savages

✨ Features That Actually Work

🛒 Add to Cart : With quantity selection and visual feedback
📝 Cart Management : View, update, remove items, clear everything
💳 Checkout Process : Multi-step form with progress indicators

🏆 Code Quality

This wasn’t some weekend hackathon code. The AI generated:

Comprehensive error handling with proper logging
Client and server-side validation
WCAG 2.1 AA accessibility compliance
Performance optimizations for concurrent users
Security best practices throughout

The final Pull Request #32 was a thing of beauty:

📁 8 new project files with complete cart functionality
🎨 15+ Blazor components for the UI
⚙ Complete service layer with interfaces and implementations
💄 Responsive styling that actually looks good
📖 Documentation explaining everything

Done in almost an hour by AI agents.

What This Means for Us Developers

Okay, so this experiment kind of broke my brain a little. Here’s what I think is happening:

🎭 We’re Becoming AI Orchestrators

Instead of writing every line of code, we’re becoming conductors of AI-powered development orchestras. The skill isn’t knowing every API by heart – it’s knowing which AI agent to use and how to coordinate them.

📋 Requirements Are Everything Now

With AI handling the implementation, the quality of your requirements becomes the main bottleneck. Garbage in, garbage out – but good requirements in, amazing code out.

🎨 Creativity Goes to User Experience

While AI handles the technical heavy lifting, we get to focus on the fun stuff – user experience, business logic, creative problem-solving.

🔍 Code Review Gets Strategic

Instead of hunting for syntax errors, we’re validating architecture, ensuring business requirements are met, and checking that the AI understood the bigger picture.

Want to Try This Yourself?

If you’re fired up to experiment with AI-driven development (and you should be!), here’s how to get started:

🛠 Get the tools : VS Code + GitHub Copilot extension
🐣 Start small : Try simple features first, build up your AI-whispering skills
📝 Write clear requirements : Seriously, this is like 80% of the battle now
🔄 Embrace iteration : Use AI feedback loops to refine implementations
👀 Learn to review AI code : Develop your “AI code quality radar”

Final stuff

Look, I’ve been coding for years, and this felt like a glimpse into a completely different future. We’re not just getting AI assistance anymore – we’re orchestrating AI agents to build entire solutions while we focus on the strategic stuff.

The shopping cart that would normally take a long time to properly plan, implement, and test? AI agents knocked it out in minutes. And it wasn’t some demo-quality code – this was (almost) production-ready, well-architected, properly tested software.

Happy coding!

Greetings