Playbook/Stage 04

Go Deeper

Level up the craft

Adding AI to Your Site

Taking a website you already have and layering AI into it, using the simplest architecture that works.

Builder’s checkBefore you add AI to anything: what is the one thing your users struggle with on your site right now that AI could actually fix? Not “what would be cool.” What is the real friction? I have built AI features that nobody asked for. They are impressive to demo and invisible in the analytics.

This section is about taking a website you already have and layering AI into it. Not rebuilding from scratch. Adding intelligence where it earns its keep, using the simplest architecture that works.

You do not need a complex stack

The architecture that works for most builders: (1) Static frontend, (2) Serverless function, (3) AI API. Three layers. You can have a working AI feature in an afternoon.

If you can build a contact form, you can build an AI feature. The pattern is identical: user fills out a form, JavaScript sends a fetch request to your serverless function, the function processes it and returns a result. The only difference is that your function calls an AI API instead of sending an email.

High-value integration patterns

  • Personalized analysis. User inputs their situation, AI applies your frameworks to give a tailored answer.
  • Interactive tools. Graders, reviewers, auditors. User pastes their work, AI evaluates against your criteria.
  • Guided exploration. User describes their situation and gets pointed to exactly the right content.

The knowledge layer

Three ways to give your AI context, simplest first: (1) Hardcoded context in the system prompt, (2) File-based context loading at request time, (3) Full RAG pipeline. Start with 1 or 2.

Set spending limits on your API accounts before you launch any public-facing AI feature. Use the cost calculator to estimate costs per user.

Builder’s checkAfter you build your first AI feature, come back to this question: does it give better answers than ChatGPT would? If not, you have a context problem, not a technology problem. The model is the commodity. Your expertise is the value.