Build a .NET 10 SaaS Template Generator with Claude Code

Two projects, two different Claude Code workflows

A developer has shared their experience building two related projects with Claude Code, highlighting distinct approaches to AI-assisted development. The first is NETrock, a .NET 10 SaaS starter template, and the second is a generator that creates customized projects from that template.

NETrock: The .NET 10 SaaS template

NETrock is an open-source .NET 10 template designed as a starting point for SaaS applications. It includes pre-wired components so developers don't have to build common infrastructure from scratch every time. Key features mentioned in the source:

Authentication system
ORM integration
Background jobs
Clean Architecture structure
Domain layer organization
Test organization patterns

The developer describes building NETrock with Claude Code as a "pair programmer" - making all architectural decisions themselves while using Claude to write code faster. They reviewed every line of Claude's output and pushed back when they disagreed with suggestions.

The generator: Client-side project customization

The generator (available at netrock.dev) lets users:

Pick a project name
Choose features from the template
Download a .zip file that builds and passes tests

Everything runs client-side - project data stays in the user's browser and doesn't hit the server. The current version is a work in progress with rough edges, and SvelteKit frontend templates aren't implemented yet.

Different Claude Code approaches

For the generator project, the developer took a different approach: "Claude in the driver's seat." They provided high-level architecture including:

How features should compose
The generation pipeline
How template markers work

Then they let Claude Code handle approximately 95% of the implementation, including:

Template engine
Dependency resolver
SvelteKit UI
Manifest system
Tests

The developer's role was primarily direction and course-correction. They encountered issues with feature combinations causing subtle bugs and template marker edge cases that required multiple iterations. The site copy also needed several rewrites because Claude "kept defaulting to marketing-speak instead of just being straight about what the thing does."

Key insights on AI-assisted development

The developer shares practical observations from their experience:

When the code is the product and you'll be maintaining it long-term, you should drive the development process or ensure you understand it deeply
When the problem is well-defined with clear boundaries, Claude can execute effectively within those constraints
The quality of your input determines which workflow works: "If your codebase is a mess, handing Claude the wheel without clear boundaries just gets you a faster mess. If your architecture is clean and your boundaries are obvious, Claude can do a scary amount of work."

The generator was built in about a week of evenings, demonstrating how quickly functional tools can be created with this approach.