This is the part I should've hit harder. The time savings argument is obvious. The lock-in problem is sneakier.
You pick a boilerplate because it has auth + Stripe + email. Three months later you realize their state management approach fights your mental model, their folder structure makes no sense for your domain, and their ORM choice means you're writing workarounds for queries that should be simple.
By then, ripping it out IS the project.
With AI generation, you can say "wait, let's restructure this" before anything is committed. You're negotiating architecture, not inheriting it.
Might turn this into its own post. Thanks for pulling that thread.
This is the part I should've hit harder. The time savings argument is obvious. The lock-in problem is sneakier.
You pick a boilerplate because it has auth + Stripe + email. Three months later you realize their state management approach fights your mental model, their folder structure makes no sense for your domain, and their ORM choice means you're writing workarounds for queries that should be simple.
By then, ripping it out IS the project.
With AI generation, you can say "wait, let's restructure this" before anything is committed. You're negotiating architecture, not inheriting it.
Might turn this into its own post. Thanks for pulling that thread.