Skip to main content

The Problem

Modern AI coding assistants:

  1. know React.
  2. know TypeScript.
  3. Next.js.
  4. ...etc.
  5. know thousands of implementation patterns.

What they do not know is:

  • Your architecture
  • Your conventions
  • Your component hierarchy
  • Your domain boundaries
  • Your engineering standards

This means that the same request can produce completely different results.

Consider:

Create a customer dashboard.

One AI session may generate:

Dashboard.tsx

Another may generate:

CustomerDashboard.tsx

Another may create:

pages/dashboard.tsx
  • Each implementation is technically valid.
  • None are necessarily aligned with your architecture.
  • AI knows how to write code.
  • The challenge is teaching AI how your organization builds software.

Mini Exercise​

Ask Copilot to create the same small feature twice, such as a customer dashboard.

Compare the two results and note:

  • file names
  • folder structure
  • component hierarchy
  • architectural consistency

Example​

One result might create:

Dashboard.tsx

while another creates:

src/pages/customer/CustomerDashboardPage.tsx
src/components/organisms/customer-dashboard/CustomerDashboard.tsx

Both may work, but only one is likely to align with an explicit architecture.