An agent that builds your lessons, not a model that writes text

Eleven at night, and a trainer is working across two windows. In one, a chatbot has produced a perfectly decent piece of writing about pricing services. In the other, the course editor inside his academy. What is left is not thinking, it is transport: a title, paragraphs, an order inside the unit, then a short quiz whose questions he will write himself from the same text. The writing was finished an hour ago. The lesson still does not exist.

The gap between "good text" and "a lesson that exists inside your academy" is what this article is about. An AI agent is not a model that writes better than the rest. The difference is that it has hands inside your platform.

Writing is not the work. Assembly is the work

Ask any trainer who has shipped a full course where the hours actually went. The answer is almost never "drafting paragraphs". The hours go into assembly: where this lesson sits, what order it takes among its siblings, which title the learner sees, which unit it belongs to, whether it has a quiz, whether that quiz is attached to it or floating somewhere unowned, and whether the sales page actually describes what the buyer will learn.

A general chatbot stops at the edge of its own window. It hands you excellent text and leaves you with the longest and dullest part of the job, the part that quietly pushes a launch back by a month. An agent crosses that edge: it writes the output and puts it where it belongs inside your academy, in the shape the platform understands. A lesson becomes an actual lesson in the course builder, with its title, content and position. A quiz becomes an actual quiz, with questions, options and answers, inside the assessments system. A page becomes a page with arranged blocks on your academy site.

That is what the word agent means in practice: not something that knows more than you, but something that executes inside the system you already work in. Which makes the right question about any generation tool not "does it write well?" but "where does it hand me what it wrote?"

What happens between your request and the draft

You write the request the way you speak: prepare an introductory lesson on time management for new managers; turn the pricing lesson into a ten-question quiz; build a sales page for the presentation skills course. No special formats, no technical commands, no magic keywords to memorise. What happens after you send it runs through four steps:

1. Deciding the output type. Lesson, quiz, assignment, educational image, or a full page? Those are five genuinely different structures, and settling the shape before writing a word is what makes the result capable of entering the platform at all instead of staying as loose text.
2. Asking you when the request is ambiguous. If your wording supports more than one reading, the agent asks before it builds rather than after. One well-placed question saves you an entire draft pointed in the wrong direction.
3. Building the whole thing. Not a snippet, not an empty skeleton for you to fill: a lesson with title, content and order; a quiz with varied questions, options and answers; an assignment with submission instructions and grading criteria; or a page with its blocks arranged.
4. Delivering it as a draft. The output is born a draft, never a publication. You open it, read it, edit it freely, then decide its fate.

Step two is the one people underrate, and it is where results diverge. You know who your learner is, what level they are at, and which example lands with them. The agent knows none of that unless you say it. The more precisely your request describes the audience and the goal, the fewer questions come back and the lighter your editing afterwards.

You can see the difference in a single line. "Write a lesson on pricing" is a request with no edges, and it returns a lesson that suits anyone and speaks to nobody. "Write an introductory lesson on pricing consulting services for freelancers in their first year, comparing three pricing methods with a worked number for each, ending in an exercise where the learner calculates their own rate" is a request with edges: defined audience, defined angle, defined ending. The second version is two lines longer and saves you half an hour of editing on the other side. That is the only skill this really asks of you, and it is closer to briefing a new assistant than to anything technical.

Approval is not an extra button. It is the design

It is easy to read "every output is a draft" as an interface detail. It is not. It is the line between a tool that helps you and a tool that implicates you.

An agent that publishes on its own means a single misunderstanding reaches your learners before it reaches you. An agent that waits for your approval means the worst possible outcome is a weak draft you discard in two seconds and nobody ever sees. The difference in probability looks small; the difference in consequence is enormous, because what has already reached a learner under your name is hard to pull back, and a teaching reputation is not built twice.

This is also why agent mode belongs to the academy owner alone. Not because your team is untrustworthy, but because "create content in the academy's name" is a different class of permission from grading or answering learners. If you distribute work across your team through team roles and permissions, treat this as the one permission that does not get distributed.

Five outputs, and how the allowance counts them

The agent runs on a renewing monthly creation allowance, counted per output rather than per word, per minute, or per size of request:

Output	What you get back	Credits
Lesson	Title, content and position in the course	1
Quiz	Varied questions with options and answers	1
Assignment	Submission instructions and grading criteria	1
Educational image	An image generated for your own material	1
Full page	A site page with arranged blocks	3

Three rules settle most questions about that table. First, the charge lands only when creation succeeds, so an attempt that produces nothing costs you nothing. Second, light edits to the draft after delivery are free, so work on it without watching a counter while you improve a paragraph. Third, the monthly allowance does not roll over into the next month, while top-up credit you buy stays until it is used.

To make the table concrete: a unit of four lessons, with a quiz at the end, one practical assignment and a landing page to sell it, comes to four plus one plus one plus three, or nine credits. That is a count of outputs, not of words. A thousand-word lesson and a three-hundred-word lesson both cost one; a ten-question quiz and a three-question quiz both cost one. Which points to a habit worth adopting on day one: ask for the whole output in one request instead of slicing it into a run of small ones, because slicing multiplies the count without improving the result.

What "does not roll over" means in practice is that this is a time budget, not a money balance. It is designed to be spent in its own month, not hoarded for a project six months out. Someone who opens the agent once a quarter pays for what they did not use; someone who builds one small unit a week gets the whole of its value.

What the agent will not do, plainly

This is the section most AI pages stay quiet about, and the one we think matters most before you set your expectations.

• It never moves first. It does not watch your academy, fill in what you forgot, or produce anything you did not ask for. Every output starts from a request you typed, so do not plan around it quietly closing the gaps in your course while you sleep.
• It knows nothing you have not told it. It does not know your learners' stories, the mistake your cohort repeats every term, or the field example that finally made the idea land. You add those, and they are usually the best part of the lesson.
• It does not guarantee every sentence is correct. Check figures, regulations and precise terminology yourself. The name at the bottom of the lesson is yours, not the model's.
• It is not on every plan. Chatting with the Allam assistant is on every plan: it answers, advises and names the screen you need. Agent mode, the part that actually creates, sits on the Premium plan.
• It does not teach for you. It produces five kinds of output, not everything. It does not run your live session, does not follow up with a struggling learner, and does not replace your judgement about who has earned a pass.

Three mistakes that make it look weaker than it is

One: asking for an entire course in a single sentence. "Build me a marketing course" is a request without edges, and the output will be as generic as the ask. The agent shines when you already know the final shape and only the building eats your time: one lesson on one defined topic for one defined audience, then a quiz built on that lesson, then a practical assignment once the unit is complete. Output by output you get a coherent course; in one swing you get a soulless draft.

Two: approving the draft exactly as it arrived. A draft is a good starting point, not a finish line. Add your example, cut the paragraph you know your cohort does not need, rewrite the opening in your own voice. Editing is free, and that is not an accounting detail: it is an explicit invitation to work the text rather than forward it.

Three: saving the allowance for the right moment. The right moment never arrives. The monthly allowance does not roll over, so the only way to benefit from it is steady use: a small unit every week beats a grand project deferred to next quarter.

Which makes the right move after reading this one request rather than a whole plan. Do not start with a course. Start with a lesson: pick a topic you know well and have been putting off for weeks, write the request in your own words to the Allam agent, then open the draft and read it as if you were the learner. Within minutes you will know where it saves you time and where your own hand is still required.

Agent mode sits on the Premium plan at 299 SAR a month, and you can try the whole platform for fourteen days with no card before committing to anything. Start from the plans page if you want to compare before you decide.