Why I built a whole platform for a course that is not finished yet
Amira and I are building a course on leadership in the age of AI. Instead of fitting it into another membership tool, I built its platform in five days with an AI agent. Why that is a sensible choice now, and what went into it.
Amira and I are building a new course: the AI Leadership Community, about personal leadership in the age of AI. The content is not finished yet. The platform it will live on is: the image above shows its program page as a member sees it. I built it in five days with an AI agent, and in this post I explain why I did not simply pick an existing membership tool.
The models are too good now
For years the sensible answer was clear: you do not build your own course platform. You rent one, or you put a plugin on WordPress. A platform needs sign-in, payments, access rules, video, mails, a member area and an admin. That was months of work for a team.
That calculation has changed. The current AI models write solid software when you give them a clear frame: automated tests, a design system, a staging server, and a person who decides. The first commit of this app is from 29 October 2026. Today the platform lists 54 features on its milestone page, has about 140 commits and runs more than 1,000 automated checks before every release. I did not type that code. I decided what it should do, looked at every screen, and said no often.
When building costs this little, the question changes. It is no longer "can I afford my own platform?" It is "do I want to bend my course around somebody else's?"
I did not want to fit into another platform
I know the existing tools well. I have run courses on WordPress with MembersSuite, I build WordPress plugins myself, and I have tried most of the hosted platforms. Each of them is good at something. Each of them also decides for you: how a lesson looks, how members sign in, which payment providers work, what the admin lets you change, and what an AI agent may touch.
For this course those decisions matter. I want the course to read like a well-made book, not like a dashboard. I want members to work through the content with their own AI assistant. I want to sell through Stripe, Digistore24 or CopeCart without moving anybody. And I want to change things on a Sunday evening without waiting for somebody's roadmap.
The features I love, in one place
Over the years I collected a list of things I liked in other tools and missed in mine. This time I could have all of them:
- Sign-in without a password. Members get a link or a six-digit code by mail.
- Lessons like in a good video course. Chapters over the video, a transcript with times, tasks, a checklist, and sources at the end like a bibliography.
- A course outline beside every lesson, with progress, the chapters of the open video and keyboard shortcuts.
- A glossary that explains itself. Terms are marked in the lessons and open a small card. Each product picks the term groups that fit its audience.
- Live dates with Zoom, a calendar subscription and the recording afterwards.
- An app on the phone, with push notices for new lessons and dates.
- Downloads and images I can replace. A new version of a handout keeps its link. A new image takes the place of the old one everywhere.
- Drafts and a trash. New lessons start as drafts, and anything deleted stays restorable for 30 days.
None of this is new on its own. What I wanted was all of it in one place, in one design, working together.
Built for working with AI
Two things make this platform different from the tools I used before.
I run it with an AI agent. The platform has its own MCP server, so Claude can create products, modules, lessons, dates and glossary terms for me. When I sketch a lesson in a chat, the agent puts it into the course as a draft. I read it, change it and publish it.
Members get the same option for their own learning. They connect their AI assistant to the course. The assistant reads the lessons they have access to, asks them questions about the content and writes notes when they ask for it. For a course about leadership in the age of AI, this was the obvious thing to offer.
How I kept it under control
Fast building is easy. Keeping the result reliable is the actual work. A few rules held it together:
- One design repository is the only source for colours, fonts and components. The app may not invent its own styles.
- Every new feature gets automated checks. A release reaches the live site only after it passes on the staging server.
- Payment keys and other secrets stay encrypted in the database. They never pass through the AI agent.
- Few dependencies: PostgreSQL, plain PHP and a little JavaScript. No build chain that breaks in a year.
All features, with the date each one arrived, are on the milestone page.
What it means for the course
The course is still in production. But I already know where every lesson will live, how members will find it and how they will work with it. The platform does not limit what the course can become. When Amira and I have a new idea for it, the platform can follow.