From Atomisation to HTML: My Honest Journey Using AI to Plan Lessons

I tried to automate everything. Here is what worked, what failed, and the free AI tool I built to help you.

We need to talk about teacher retention.

I have long campaigned for the idea that schools should provide high-quality, pre-planned resources. Teachers should be freed up to focus on the art of teaching, the delivery, the relationships, the responsive coaching, rather than spending their evenings reinventing/improving the wheel on a PowerPoint slide.

The burden of resource creation is driving trainee teachers to quit before they start being teachers and pushing experienced educators out of the profession. We are losing great people because the workload mathematics just doesn’t add up.

Inspired by the work of experts like Craig Barton (whose Tips for Teachers and Mr Barton Maths podcasts are essentials), I have been on a mission to streamline my workflow. I wanted to see if AI could be the lever that finally fixes the planning burden.

Over the last year, I have been experimenting with a hybrid workflow: combining rigorous pedagogical research with the power of Google Gemini.

Here is the update on my journey so far, the wins, the “not-yets,” and the tools you can use today.

Step 1: The Foundation (Atomisation)

It all started last year when I discovered the concept of Atomisation, pioneered by Kristopher Boulton (CEO of Unstoppable Learning).

If you aren’t following Kris on Substack or reading his work, you should be. The premise is breaking down complex learning objectives into their smallest, accessible “atoms” that can be chained.

This approach changed my teaching. It allowed me to create lesson sequences where lower-attaining students could make the same progress as high-attainers, simply because the scaffold was so precise. However, doing this manually is cognitively demanding. It takes time to break a concept down that finely. I needed a way to speed this up without losing the quality.

Step 2: The “Gemini Gem”

This was the game-changer. I realised that while AI struggles with “creativity,” it excels at “structure.”

I created a custom Gemini Gem (a custom version of the AI) specifically trained on Boulton’s principles of Atomisation. I fed it the pedagogical rules, and suddenly, I could input a topic, and the AI would do the heavy lifting of breaking it down for me.

Note: I have made this Gem available for free. You can click the link at the bottom of this post to try it out.

Step 3: From Atoms to Lesson Plans

Once the AI could “atomise,” the natural next step was the lesson plan.

I upgraded the Gem. Now, I don’t just ask for the atoms; I ask for a full lesson architecture. I prompted the AI to generate:

• Teacher scripts (for explicit instruction).

• Detailed activities based on the atoms.

• Plenaries.

This works brilliantly. It removes the “blank page syndrome” and provides a rigorous pedagogical skeleton for the lesson in seconds.

Step 4: The Revelation (The “Slide Deck” Wall)

Here is where I hit a roadblock.

My ultimate dream was to have the AI take that text-based lesson plan and automatically generate my Google Slides, using my specific school templates.

I know this is possible. I have seen brilliant colleagues and coding experts share methods using Marp or specific AI plugins to generate slides. I tried several of these workflows.

My honest verdict? Right now, for me, the juice isn’t worth the squeeze.

I found that debugging the code or tweaking the AI-generated slides took just as long, if not longer, than making the slides manually. Teaching is a visual medium, and AI currently struggles to lay out visual information as intuitively as a human can.

I have decided to pause on automating slides. And that is okay. It’s not a failure; it’s a pivot. I am sticking to manual slide creation for now, knowing that high-quality slides are reusable assets that I only have to build once.

Step 5: The Breakthrough (Worksheets & HTML)

Since slides were too time-consuming to automate, I turned my attention to independent practice.

This has been a massive success.

I started by asking my Gemini Gem to write code for LaTeX (a typesetting system used for high-quality math and science documents). I can paste this code into an editor like Overleaf, and it generates professional, textbook-quality PDF worksheets instantly. I can ask Gemini to “make Q4 harder” or “add a scaffold to Q2,” and it rewrites the code in seconds.

But the real leap forward came recently, inspired by a Substack post from Harry Zafar.

Harry shared a method for creating HTML worksheets. Instead of a static PDF, the AI writes a simple webpage.

• It looks clean and organised.

• It can be opened in any browser.

• Crucially, you can code it so that answers are revealed when students (or the teacher) click on the question.

Harry also wrote another Substack, detailing how to use this for Checking for Understanding, generating specific questions for Mini-Whiteboard sessions.

Where I am now

I have a workflow that works:

1. AI handles the cognitive heavy lifting (Atomisation and Planning).

2. I handle the visual storytelling (Slides).

3. AI & Code handle the resource generation (HTML/LaTeX Worksheets).

This is the most sustainable my workload has felt in years.

If you want to try the Gemini Gem I built for Atomisation, click this link. And if you’re interested in a tutorial on the HTML worksheets, let me know in the comments or join the subscriber chat!