Learning Designer Agent: AI-Powered Course Generator
Builds a complete course with you, in a conversation. You describe what you need, the agent asks one question at a time, shows you what it will produce, and only generates after you confirm. From a PDF or from a plain briefing.
What it does
The Learning Designer Agent, also known as Gi Learn, turns a document, a manual or an idea into a structured course ready to review and publish. It is part of GFoundry Intelligence.
The difference from a one-shot generator is the conversation. You brief a designer that asks what it needs, tells you the limits before you hit them, and hands back a summary for approval.
Key capabilities:
Conversational briefing: one question at a time. It proposes options and always accepts a free-form answer.
You choose the blocks: a structured content page is always included; on top of that you pick at least one of slides, a quiz, a downloadable PDF and a curated YouTube video, or all four.
Control of the content: counts, named sections, section types, topics to focus, topics to exclude, tone, level and sentences that must appear word for word.
Your document as the source: upload a PDF and decide whether the course may only use that document, or may also draw on the model's general knowledge.
Confirmation before generating: it summarises the final numbers and asks before generating.
No silent trimming: ask for more than a block allows and it tells you the limit and offers the maximum or splitting the content into more than one course.
Use cases
Onboarding, compliance training, upskilling, product training, policy rollout. Whenever you need to turn an internal source, or your own subject knowledge, into a structured course.
It is particularly useful when the course has to say something specific. A compliance course where six named principles must each appear with an explanation, an expected behaviour and an example is exactly the kind of brief the conversation is built to capture.
How do I create a course?
In the backoffice, open Learn, select Create Learn Content and choose Generate Content with AI. The Learning Designer Agent opens.
To work from a document, use Upload PDF and then Start with this PDF. PDFs can only be attached at the start of the conversation, and the limit is 50 MB.
Describe the course you want, in your own words. A sentence is enough to start.
Answer the agent's questions. It needs a course title, a course objective (one sentence saying what the course lets someone do) and a target audience, plus which additional blocks you want.
Go deeper wherever it matters to you: topics, exclusions, tone, level, structure, counts. Skip whatever does not, and the agent decides.
Read the summary. It lists the blocks and the final counts.
When it asks Can I generate the course based on this?, answer yes. Generation starts in the background.
You can delegate any of the three required fields. Say "you decide" or "suggest one" and the agent writes it, shows it to you in the summary, and does not ask you again.
What can I decide in the conversation?
Everything below is optional, and anything you leave out the agent decides for you.
Which blocks: the structured content page is always generated. Choose one or more of slides, quiz, downloadable PDF and a curated YouTube video.
How much of each: quiz 1 to 50 questions, multiple choice and/or true or false, with 2 to 4 answer options per question (3 by default; four is the most the platform stores). Slides 1 to 60 content slides. Structured content page 1 to 12 sections. Downloadable PDF 1 to 6 modules. Practical scenarios 1 to 8. One curated YouTube video.
Cards in the slides: on top of the content slides, the deck includes Did you know? cards (0 to 12, 4 by default) and True or false? cards (0 to 10, 3 by default). The summary shows the deck total, for example 20 content slides plus 4 and 3 cards make 27, and you can set the cards to zero.
The structure of the content page: name the sections yourself, in your order, with your titles. The agent keeps them as you gave them. If the document has more than 12 top-level sections, the agent proposes grouping them into at most 12 chapters.
The shape of each section: text, callout, highlight, definition list, comparison, card grid, steps, FAQ, checklist, list, table or scenarios.
Items that must appear: dictate a list and it is carried over word for word. Ask for one example per category and each category gets its own example, not a bare list of names.
Grouped content: ask for two columns, such as benefits against risks, or for six groups each with an explanation, an expected behaviour and an example, and the course is built that way.
Exact sentences: give a sentence that must appear verbatim, for example a line from a policy, and it is reproduced literally in the section you chose.
Learning objectives: dictate them and they are kept word for word rather than reworded.
Tone and level: tone professional, neutral or casual. Level beginner, intermediate or advanced.
When you give a number, the agent records that number. When you give a range, the agent picks a value inside it and shows it in the summary, so you can adjust it before confirming.
How do I work from my own document?
Upload the PDF and the agent summarises it back to you, then confirms how strictly the course should stick to it. There are three source scopes:
Document only: the course uses nothing but the uploaded document. This is the right choice for compliance, policy and contractual content, where outside facts are a liability.
Document plus knowledge: the document leads, and the model fills genuine gaps.
Knowledge only: the course is written from the model's subject knowledge. This is the scope when there is no document.
With a document uploaded, document only is the default, so you have to opt in to anything looser.
How do I keep a topic out of the course?
Exclusions come in three kinds, because "leave this out" rarely means the same thing twice. Say which you mean and the agent records it accordingly:
Must never appear: the topic is kept out of the course entirely.
Covered in a later module: at most a short note that it comes later. No section, no definition, no example.
Mention is fine, do not develop: the topic can be named in passing but never expanded.
Be explicit about which kind you mean.
What is generated?
The course arrives in the Learn module, ready for review. Depending on what you asked for:
Title and description, plus skill tags for the course.
Structured content page: the core block, built from the sections and section types agreed in the conversation. Use it as an introduction, as the first course content or as a developed information block.
Slides: with the Did you know? and True or false? cards.
Quiz: every question has one correct answer.
Downloadable PDF: a document organised in modules, with text, tables, lists, concepts, tips, warnings and quotes.
Curated YouTube video: chosen for relevance to your topic, objectives and audience, and for source credibility, quality, language, length and how current it is. It is an existing video selected for you, not a video the system produces.
Practical scenarios: rendered inside the course content rather than as a separate block.
Generation runs in the background and takes a few minutes, depending on how many blocks you asked for and how large the source document is. Select Continue in background to keep working; you can check progress by reopening the Learning Designer Agent. When it finishes you see Your course is ready for review. and Open course.
What if one block fails?
Blocks are generated independently, so one failure does not lose the whole course. When that happens the course is still created, and the agent shows Some blocks couldn't be generated with the blocks that failed. Review the course and add those blocks manually.
If generation fails as a whole, you see Course generation failed. Please try again. instead of an empty course.
In which language is the course generated?
In your platform's base language. The agent writes the title, audience, objectives and section titles in that language even if you talk to it in another one. The exception is wording you dictate explicitly, such as an exact title or a sentence that must appear: it is kept exactly as you gave it, in the language you gave it.
If your account has other languages configured, the course shows AI translations pending, with two options:
Generate Translations: generate the other languages with AI. It can be done only once for each course, and editing is paused while the translations are generated. Validate the base version first, so you do not have to correct the same problem in every language.
I don't want to use AI generation. I'll add my own translations.: fill the other languages with the base language content, so you can edit them yourself. This cannot be undone.
Publishing stays disabled until one of the two is completed. You can still send the course for review in the base language.
Can I edit the generated content?
Yes, and you should. Every part of the course can be reviewed and edited before publishing.
Treat the output as a strong draft, not a finished course. AI models make mistakes, so read the text and look at the generated images before you publish. Use the review pass to add company-specific examples and to check anything with legal, compliance or contractual weight.
Tips for better courses
Answer the objective question properly: "Recognise risks and apply safe behaviours at work" produces a far better course than "learn about AI". The objective steers every block.
Say what must appear: if the course exists to cover seven specific concepts, name the seven.
Upload the document for strict scope: for policy and compliance content, upload the PDF with Start with this PDF and keep document only.
One course at a time: the agent generates a single course per run. If you are planning a path of five modules, brief them one by one.
Give the exclusions: saying what to leave out shapes a course as much as saying what to include, especially when another module already covers it.
Correct at the summary, not after: the summary before confirmation is the cheap place to fix a count or a section title.
Is there a usage limit?
If your platform has a quota for the Learning Designer Agent, the agent window shows how much of it you have used, for text requests and for file uploads.
Privacy and data handling
The Learning Designer Agent uses an external AI provider (OpenAI) to draft content from your briefing and your uploaded document. Generated content is stored in your GFoundry Learn module like any other Learn Content, and uploaded documents are scoped to your organisation. The Data Processing Agreement, including the list of sub-processors, is available on request.
Which article should I open next?
To edit, improve or translate any generated text field, see AI Writing Assistant: generate, improve, shorten, fix and translate text in backoffice forms.
For the same steps from inside the Learn module, see Creating Learn Content with AI.