The Summer of Code, update #3

Almost two weeks ago I published The Summer of Code, update #2.

I have been quiet since then.

Not because I stopped building. Quite the opposite.

I did enough work for several updates, but most of it is not public yet. I don’t want to announce projects before they are ready, so this update will be a little vague on purpose.

I am working on a ton of things. I’ll publish them soon.

I started a lot of new projects

Probably too many.

Over the past two weeks I went from an empty folder to a working product several times.

Some are tools. Some are content-heavy sites. Some are small products built around one problem. Others started small and quickly became much bigger than I expected.

Together they now include hundreds of interactive pages, large libraries of original content, checkout flows, emails, accounts, dashboards, downloads, and production deployments.

I built the first version, used it, found what felt wrong, and rebuilt it. Sometimes I removed half of it. Sometimes I changed the entire direction after seeing the real thing in the browser.

This is one of the biggest changes AI agents brought to my work.

I don’t have to protect an idea for months before I can see it. I can build a real version, test how it feels, and decide if it deserves more time.

Building became cheaper. Judgment became more important.

Not everything will become a public product. That’s fine. The goal of the Summer of Code is to follow ideas and see where they go.

I built and rebuilt courses

I also did a lot of work on my educational products.

I rebuilt one curriculum from scratch, created a new 20-lesson program, expanded another course with new guides and tutorials, and reorganized how all my programs fit together.

This last part was harder than writing lessons.

I had several good courses, but the path between them was not clear. A beginner, an experienced developer, someone with a product idea, and someone building an independent business need different things.

I spent time defining those boundaries. What each program teaches, who it is for, what it does not cover, and where you should go next.

That work is now visible on the courses page and on the homepage.

The goal is simple: tell me where you are, and I can point you to the next useful step.

I wrote more than 20 tutorials

One day I decided to work on the tutorial queue.

That turned into more than 20 new long-form technical tutorials.

They cover testing, browser extensions, APIs, observability, background jobs, package development, desktop apps, deployment, AI infrastructure, and more.

Most are scheduled to appear gradually over the next few weeks. I don’t want to publish all of them on the same day and flood the blog.

I also refreshed more than 100 older tutorials.

Old technical posts decay. Commands change, tools disappear, links break, and advice that made sense years ago becomes confusing.

The refresh removed obsolete material, checked claims against current documentation, and made the useful posts useful again.

This kind of maintenance is not as exciting as launching something new, but with a site this large it matters a lot.

I kept improving the AI Workshop

The AI Workshop now has its own home.

I moved the content into the site, rebuilt the public pages, added a complete changelog, and published new material on open models and multi-agent workflows.

I also simplified the relationship between the workshop and my other programs.

The workshop is for people who can already build software and want to go deeper into AI. It covers the tools and complex workflows a developer should know: agents, models, context, skills, MCP, sandboxes, reviews, and automation.

It is not a coding course from zero. Making that clear helps everyone choose better.

I changed how I build

I now start new web projects with the same default stack: Astro, HTMX, and Alpine.js, deployed to Cloudflare.

Having a default removes a surprising amount of friction.

I don’t spend the first day choosing a framework, a hosting platform, or a client-side state library. I start building the product.

AI agents do most of the implementation, but I stay very involved. I define the direction, review the work, use the result, reject weak ideas, and decide what gets removed.

The fast part is producing code.

The real work is deciding what should exist.

More soon

This is probably the least detailed Summer of Code update so far.

That is deliberate.

There are several projects I am excited to show you, but I want each one to have its own proper introduction when it is ready.

For now, the short version is this: I did a ton of stuff, I am working on a ton of projects, and the Summer of Code is in full swing.

I’ll publish them soon.

Related posts about news: