This guide exists as a website rather than a PDF because AI tools made it practical. Not in theory -- in actual practice, across about four days of part-time work, using a combination of Claude, Cowork, Perplexity, and a hosting platform called Netlify that most people have never heard of.

This page is the honest story of how that happened. What worked, what did not, what I explored along the way, and what the whole process illustrates about using AI tools in the real world.

Where It Started: A Word Document

The guide began as a Word document. That is where most writing begins. The first version was a linear manuscript -- front to back, like a book -- covering the AI tools I use, what they are good for, and how to get started.

The problem with a Word document is that it is a terrible format for content that changes constantly. Pricing changes. New models appear. Tools get acquired or shut down. Every time something changed, I had to open the document, find the relevant section, update it, and redistribute the file. There is no link to share. No way to update it for everyone at once. No navigation for readers who only want one part.

The obvious answer was a website.

Exploring the Options

My first instinct was GitBook -- a platform designed specifically for documentation and guides. It has a visual editor, free hosting, and produces clean, readable output. I asked Claude to convert the manuscript into GitBook's markdown format, which it did: individual markdown files, a table of contents, navigation structure, the works.

I spent time exploring GitBook and it has real merit. But after working through it, I decided the setup complexity was not worth it for what I needed right now. GitBook is designed for teams maintaining technical documentation. I am one person maintaining a guide. The mismatch was subtle but real.

So I went simpler.

Building the Site With Claude

Claude built the site as static HTML -- one file per section of the guide, plus a shared CSS stylesheet. No database. No server-side code. No build tools. Just files that any web browser can open directly.

It did this by writing a Python script that converted the manuscript content into HTML pages. The script generated a consistent navigation sidebar on every page, previous/next page links at the bottom of each section, properly formatted tables with italicised prompt examples and bold tool names, a mobile-friendly layout with a hamburger menu on small screens, and clean typography using Source Serif for body text and DM Sans for headings.

This took one conversation. I described what I wanted, Claude wrote the generator script, the script produced the HTML files, and the site was ready to go.

The result was not perfect -- nothing ever is on the first pass -- but it was usable within an hour of starting.

Hosting on Netlify

The hosting platform is Netlify. It has a feature I had never encountered before: you literally drag a folder onto their website and your site goes live. That is it. No command line. No configuration. No server to manage. Within about 30 seconds of dragging the folder, the site had a URL anyone could visit.

The free Netlify tier handles it comfortably, which means the hosting cost is zero.

For a site like this, it has been remarkably easy to implement. I say that as someone who is not a web developer and had never deployed a website before starting this project.

The Editing Workflow

The files live on my computer. When I want to update the site -- fix a price, correct a tool description, add a new section -- I edit the relevant HTML file locally, then drag the updated folder back to Netlify to redeploy. It takes about 30 seconds.

That is the entire workflow. Edit locally, drag to Netlify, done.

I considered connecting Netlify to a GitHub repository, which would mean changes deploy automatically the moment I save a file, without the drag step. But honestly, 30 seconds is not a problem worth solving right now. The drag-and-drop approach is simple, it works, and it keeps the workflow entirely in my control without depending on another platform.

Using Cowork for Bulk Updates

Claude's Cowork feature changed how I handle large-scale updates to the site. Rather than editing files one at a time, Cowork lets me point it at the folder of HTML files on my computer, describe what I want done across all of them, and let it work through the task autonomously.

A practical example: when pricing needed updating across multiple tool profiles -- Claude, ChatGPT, Gemini, Perplexity -- I did not update each page individually. I used Cowork to find every instance of the old pricing information across all HTML files and update them consistently, then dragged the revised folder to Netlify.

Cowork is also useful for the structural tasks that would be tedious to do manually: checking that all internal links still work, ensuring consistent formatting across pages, or reorganising content when the structure of the guide changes.

It is not perfect -- complex edits sometimes need a review pass before deploying -- but for bulk updates across many files, it saves significant time.

The Top 100 AI Tools Page

One of the more ambitious additions to the site is a Top 100 AI Tools page -- a ranked list of the most useful AI tools, with pricing, brief descriptions, and links.

The challenge is that this list changes constantly. Tools rise and fall. Pricing updates. New tools emerge from nowhere and become essential in weeks. A static page written once and updated manually would be out of date almost immediately.

The solution is a scheduled agent that runs weekly and keeps the list current automatically.

The data layer. The tool data lives in a structured JSON file. Each entry includes: name, category, pricing, free tier details, brief description, link, and a last-verified date. This is the single source of truth for the page.

The weekly research pass. A scheduled Claude agent runs every Sunday. It uses Claude's research and web search capabilities to check pricing against current websites, look for new tools worth including, flag tools that have been discontinued or significantly changed, and identify any ranking that should shift based on current adoption.

The agent produces a summary of proposed changes rather than applying them directly.

The human review step. I review the proposed updates, approve or adjust them, and redeploy. The whole review typically takes ten minutes. I have kept a human in the loop deliberately -- AI research still makes mistakes, and some judgements about ranking and inclusion are not purely objective. Those stay mine.

The Multi-Tool Workflow

What this project illustrates, more than anything else, is how AI tools work in combination. No single tool did everything.

Claude was the primary building tool. Site generation, HTML file updates, structural decisions, and the content itself.

Cowork handled the bulk tasks -- updating multiple files at once, running consistency checks across the site, managing the repetitive parts of keeping 50+ pages coherent.

Perplexity was my research tool for the Top 100 content. When I needed to verify tool capabilities or understand where something stood in the market, Perplexity gave me sourced answers I could verify quickly.

Netlify handled hosting and deployment -- drag, drop, live.

The scheduled agent runs the weekly research pass and keeps the Top 100 current.

None of these are complicated individually. The power comes from connecting them into a workflow where the tedious parts happen automatically, and the judgement calls stay with me.

What This Demonstrates

This whole project is a live example of the guide's core argument. But let me be honest about where I started, because it matters.

I have worked in IT for a long time. I programmed computers early in my career. For the past 15 years I have been entirely business-focused -- in non-technical roles, though always technology-aware. I have deployed websites before: my own business site, built on Squarespace, plus time spent with Substack and a couple of other point-and-click builders. I know my way around those tools. But they are a completely different category from what I have done here. Squarespace gives you templates and a drag-and-drop editor. This involved Claude generating HTML files from a Python script, a CSS stylesheet, and a hosting platform most people have never heard of. The gap between those two experiences is significant.

That context matters for two reasons. First, it means this is not a story about a technical person doing a technical thing. Second, it means if you are completely new to IT in general, there may be a steeper learning curve than I experienced. I will not pretend otherwise.

What I will say is this: the way we interact with these tools now -- natural language, photos, voice, conversation -- has fundamentally changed what is accessible. You do not write code. You describe what you want. You iterate. You ask questions. The tools meet you where you are. Once you get your head around the basics of the toolset and how each piece fits together, the barrier drops dramatically. The hard part is no longer technical. It is knowing what you want to build and being willing to work through the early confusion of learning a new workflow.

Four part-time days after starting, I have a 50+ page website with a mobile-friendly design, a scheduled updating agent for the Top 100 page, and a workflow simple enough that I can update and redeploy any page in under five minutes.

That is the point. Not that AI replaces the thinking. That it handles enough of the mechanics that the thinking is all you need to bring.

Built in four part-time days starting March 2026. The site continues to evolve.