What this article is about
If you're just starting to learn AI, it's easy to fixate on "Can AI make me a nice-looking presentation?" What this class really demonstrates is that every round of prep, organizing, editing the slides, and tuning the style can be recorded and turned into a system the next Agent can understand and pick up. As of 2026-07-16, this article also mirrors the formal SOP in the main knowledge base: "Linked SOP: The Lecturer's Slide-Making Agent Workflow."
· People who can already write with ChatGPT or Claude but aren't sure what more a desktop Agent can do
· People who keep redoing slides, restating requirements, and re-fixing formatting
· Anyone who wants to turn their teaching, consulting, research, or content-production process into a method AI can run again and again
· A view of a complete Agent workflow, from gathering data to producing slides
· Why you should produce a text outline first and only then build the HTML slides
· How to put your preferences, process, checklists, and examples into a folder so the next run can carry on
Making slides is just the entry point; the real focus is the workflow
This class makes its point clear from the start: making the slides, doing it with an Agent, and understanding how to put your own knowledge into a knowledge base.
On the surface, this is a class about using AI to make slides. In practice, it breaks a routine task into a repeatable process.
First have the course theme or original syllabus, and then collect the pre-course questionnaire, unit needs and student questions.
Put the questionnaire and the course syllabus into the same folder, and first ask the Agent to report which files it has read.
Ask the Agent to organize a text framework of ten pages or fewer, and confirm the direction before moving on to building the slides.
Once the direction is right, ask the desktop Agent to produce HTML slides and keep the file in the project folder.
Write down preferences such as font size, color scheme, tone, layout, and folder location.
Write the repeatable process, examples, and self-check list into operating rules the Agent can load next time.
Step 1: Put the questionnaire and course syllabus into the same folder
The real lesson preparation method demonstrated by the teacher is to first have the course theme or original syllabus, and then collect student questionnaires or unit needs. When the information comes in, the first step is to put the syllabus, questionnaire, and demonstration materials into the same project folder.
This action is very important because the Agent needs to know which batch of data it is processing. It is best for newbies to open a practice folder first and not let the AI process the really important files at the beginning. After the process is stable, put it into the official work folder.
Let the Agent report the contents of the folder first to confirm that it can read the data and to confirm that it has not missed any key files.
The second step: create a text outline first, and then make a beautiful presentation
After confirming that the Agent has read the course syllabus and questionnaire, the next step is to ask it to organize the presentation outline.
The focus of this step is to first align the content direction. What students really care about and what the original syllabus wants to talk about must be connected first. If AI is required to create a beautiful presentation from the beginning, errors will be included in the layout, making it more difficult to correct later.
Confirm with a text outline first, and you can spread out the issues: which issues should be discussed in advance, which content can be merged, which cases should be replaced with a version that this unit can understand, and which pages have too many and should be deleted first.
Step 3: Understand the difference between web chat AI and desktop work AI
Web version AI is very suitable for reading information, organizing outlines, and helping you think about page content. But if you want to write files directly in the local folder, generate HTML, save versions, and leave files in the project, the desktop Agent is more suitable.
Suitable for discussing content, organizing ideas, and producing a first draft. The main limitation is that after output, it often needs to be manually downloaded, pasted, and moved to other tools.
Suitable for reading local folders, writing files, editing files, and producing finished work you can save and reuse.
A desktop Agent like Codex is basically a desktop version of GPT installed on your computer. It can step out of the web chat box and, in the folder you specify, organize files, generate files, edit web pages, and leave work results behind.
Step 4: Write preferences into the folder
After many people use AI to give a presentation, they still have to give it again next time: the font is too small, the color is wrong, the page is too colorful, the tone is too strong, and the title is too like a marketing copy.
This lesson reminds you that just because AI says "I remember" does not mean that you are really in control. If a preference only exists in a certain conversation, a certain AI tool, or a certain computer, it may disappear if the tool or situation is changed.
A more stable approach is to write the preferences in the project folder. For example, font level rules, color preference, commonly used layouts, course tone, target audience differences, and work diary after each execution.
Step 5: Make repetitive processes into skill packages
When you find yourself saying the same thing over and over again, consider turning it into a skill package. The process demonstrated in this class is very suitable for encapsulation: organize the questionnaire, integrate the syllabus, produce a text outline, make HTML after confirmation, apply a fixed style, and self-check after completion.
The prompt word is just a current instruction. The skill package writes down fixed procedures, preferences, examples, and inspection rules so that the Agent knows which method to follow in similar situations.
When receiving the course syllabus and questionnaire, you must first confirm the content and source of the folder.
You cannot directly develop the final product. You must first make a text draft and then make HTML after confirmation.
There should be a self-check list at the end of the process to prevent the model from skipping critical steps.
Step 6: Accumulate the products after each class into a knowledge base
After a course is over, the presentation is not the only thing worth staying with. There are also pre-class questionnaires, course outlines, presentation outlines, finished HTML products, style preferences, student feedback, post-class QA, and problems discovered after this completion.
If these materials are kept in the same traceable location, the Agent can be asked to look back during the next lesson preparation: how the same topic was taught last time, where students get stuck most often, which metaphors are easy to use, which pages are too difficult, and which styles are suitable for this audience.
When accumulated to a certain extent, this folder will grow into a knowledge base that Agent can read, connect to, and help you do things for you.
Official SOP version: Lecturer giving presentation Agent workflow
This process currently has two landing points: the public page you are looking at is responsible for allowing readers to understand the concept; the SOP in the main knowledge base is responsible for allowing Agents, skill packages and subsequent tasks to reference the same operating rules.
Concentrate the course syllabus, pre-class questionnaire, old presentations, hosting requirements and supplementary materials, and first let the Agent report what it read.
Group students’ questions and compare them with the original syllabus to determine which are core, required, supplementary and extended questions, or QA questions.
Confirm the page skeleton, teaching rhythm, and purpose of each section first, so you don't bake mistakes into a polished layout.
Once the direction is confirmed, use the desktop Agent to produce HTML slides, a PDF, or a PowerPoint draft.
Write the fixed process, layout preferences, font-size rules, checklists, and examples back into the lecture-prep skill package.
Keep the original materials, finished work, preferences, pitfalls, student feedback, and the SOP you can reuse next time.
How to link skill package and knowledge base
This public page is the "entrance for people to see"; the main knowledge base _agent/skills/lecture-prep/references/ is "the official document referenced by the Agent". The contents of the two are aligned, but their uses are different.
Handles the execution side: when it receives the syllabus and questionnaire, it reminds the Agent to take stock of the materials first, produce a text outline first, then build the finished slides, and finally self-check.
Handles the memory side: it keeps the process version, inputs and outputs, checkpoints, pitfalls, and source links, so the next prep task doesn't have to re-explain the background.
In other words, the skill package determines "what the Agent will do this time", and the knowledge base determines "whether it can handle it next time." If you only have the finished brief, you will still start from scratch next time; if the process, preferences, and checklists are all saved back, you can accelerate on the same path next time.
Source and Navigation
- Public source:Turn the pre-class questionnaire into a presentation, and then save the process into a skill package
- Extended reading:How lecturers use knowledge management processes to turn lesson preparation into a reusable system
- Skill package hub:Coach Jiang skill package downloads
- Main knowledge base version:
_agent/skills/lecture-prep/references/2026-07-16-2314 Linked SOP: Lecturer giving presentation Agent workflow｜Lecturer, presentation, AIAgent, skill package, knowledge base.md
Last updated: 2026-07-16 23:18 CST. Restrictions not yet completed: This page has been synchronized to the official website source code; if the social platform still displays the old summary or old cover, it is usually the platform cache, and you need to repost the link or wait for the cache to refresh.
How you can practice
Don’t experiment with the most important information first. Open an exercise folder, put a task description or course syllabus, and then put a questionnaire, interview record or demand summary.
- First ask the Agent to report what is in the folder.
- Compile a text outline based on the data.
- When you’re done, write down your process, preferences, and checklist.
If you will do similar tasks next time, organize this process into a skill package. Once you've done it three to five times, you'll start to see your knowledge base take shape.