Using Agentic AI To Create And Refine Teaching Resources

Download the resources

• Download the lesson plan (Word document)
• Download the PowerPoint presentation

Starting with the teaching need

The starting point was not “make me a lesson”. It was a specific teaching need.

I wanted a one-hour lesson for a Year 10 group studying OCR GCSE Computer Science. I knew the specification area, the time available, the ability range of the class, and the tools students had access to. They could use Trinket on their computers, but I also wanted paper-based tasks because trace tables are often better understood when students physically write out the changing values.

That detail mattered. It meant the AI was working within a real teaching scenario rather than producing a generic lesson outline. The same principle applies in other settings: a university seminar, a workshop for apprentices, a short course for adult learners, or a professional development session all need context before AI can be genuinely useful.

The first version gave me a structured lesson with a starter, teacher input, student activities, trace table examples, a Trinket debugging task and a plenary.

That first draft was useful, but it was only the beginning.

Building the resource in-situ

What made the workflow particularly useful was that Cursor could create the actual files in place. Instead of copying text from a chat window into a document, I asked it to create the lesson plan in my Documents folder.

The initial version was created as a Markdown document. I then asked for it as a Word document, and Cursor converted it into a .docx file. After that, I asked it to format and design the document more carefully. It rebuilt the file with a clearer structure, styled headings, colour-coded sections, tables, activity boxes, code examples, trace tables, differentiation notes and teacher guidance.

This is where the agentic element becomes valuable. The AI was not just generating text. It was carrying out a sequence of practical tasks:

• creating the document
• converting it into the required format
• improving the layout
• validating that the file opened correctly
• making further adjustments when asked

That saved time, but it also kept the work close to the real output. I could judge the actual lesson plan as a usable teaching document rather than only reading a response in a chat.

This distinction is useful for educators beyond schools. A lecturer might use the same workflow to create a seminar handout, lab worksheet or assessment briefing. An FE tutor might generate a workshop plan and learner activity pack. An adult education provider might produce a session plan, slides and follow-up materials for learners returning to study.

Creating the presentation

Once the lesson plan was in place, I asked Cursor to create a PowerPoint presentation to go with it. Again, this was not simply a case of “make slides”. The first version gave me a complete teaching deck with slides for the starter, objectives, reference language, matching activity, common errors, trace tables, Trinket debugging and plenary.

Then came the part that feels most like real teaching resource development: refinement.

When I opened the presentation in LibreOffice, some of the formatting was not quite right. Some text did not fit comfortably in the boxes. I fed that judgement back into the process and asked Cursor to adjust it. It regenerated the deck with fewer crowded slides, simpler layouts, larger boxes and better spacing.

Later, I decided that the worked FOR loop example and its trace table should be on the same slide. From a teaching point of view, that made sense: students need to see the algorithm and the trace table together so they can connect each line of the algorithm to the changing variable values. Cursor updated the PowerPoint accordingly and removed the now-redundant separate solution slide.

Finally, I noticed that although the presentation looked fine in LibreOffice, it was a little off when imported into Google Slides. That is a very normal problem when moving presentation files between tools. I asked Cursor to optimise the deck for Google Slides. It simplified the formatting, used safer fonts such as Arial and Courier New, removed elements that were more likely to import awkwardly, and kept the layout clear.

The role of professional judgement

The important point here is that the educator remained in control of the process.

The AI could generate, format and adjust the resources, but the decisions about what worked pedagogically came from professional judgement. For example:

• The lesson needed both Trinket and paper-based work because the class had a range of abilities.
• Trace tables needed enough space and clarity for students to follow the logic.
• The FOR loop example and trace table belonged on the same slide because they support each other visually.
• The deck needed to work in Google Slides because that is a realistic platform for future use.
• The final resource had to be classroom-ready, not just technically correct.

This is where I think agentic AI is especially powerful for educators. It can handle much of the production work, but the educator still directs the resource based on knowledge of the learners, the curriculum or programme, the learning objective and the practical constraints of the session.

What the final lesson includes

The finished lesson is designed for a mixed-ability Year 10 group and includes:

• a starter activity based on spotting an error in an algorithm
• a comparison of OCR reference language and Python
• a matching activity to connect reference language with Python syntax
• explanations of syntax, logic and runtime errors
• worked and independent trace table tasks
• a Trinket debugging activity
• extension tasks for higher-attaining students
• an exit ticket plenary

The lesson aims to help students see algorithms as something they can inspect, test and refine. That is valuable both for programming and for OCR exam questions, where students are often asked to read, trace, correct or write algorithms.

Why this workflow matters

For educators, time is often the limiting factor. Many of us know what we want a session to do, but producing the supporting materials can take a long time: lesson plans, slides, activities, worksheets, differentiation, formatting, file conversion and platform compatibility.

Using Cursor in this way made the process more fluid. I could move from idea, to lesson plan, to Word document, to presentation, to revised presentation, to Google Slides-optimised version without leaving the workflow. Each step was shaped by feedback and professional judgement.

That is a different model from simply asking AI for a lesson plan. It is closer to working with a capable assistant who can create, edit, convert and refine materials while I focus on whether the resource will actually work for my learners.

This is also why organisations such as Terrabase have a role to play in education. The value is not just in knowing that AI tools exist, but in helping educators, departments and training providers identify realistic workflows where those tools can make a practical difference.

Final thoughts

Agentic AI will not replace the expertise of teachers, lecturers, tutors or trainers. The value comes when it amplifies that expertise. In this example, Cursor helped build a complete lesson package quickly, but the direction came from the teaching need and the refinements came from professional judgement.

The process was iterative: generate, inspect, improve, test, adjust and optimise. That is exactly how good teaching resources are often made, but with much of the repetitive production work handled by the AI.

For me, that is the most compelling use of agentic AI in education: not replacing the educator, but giving educators a faster and more flexible way to turn professional judgement into high-quality learning materials.