What can the AI agent create for my users?

Every AI creation route in Coassemble, what each one does, and what to check before promising one to a customer.

Course creation runs as a conversation. Your user brings what they have, answers a few questions, and watches a structured course get built, rather than choosing a creation path up front and then deciding separately how their source material should be handled.

1. Bring the material. Upload a Word document, PowerPoint or PDF, upload a SCORM package, or describe the course in plain language.
2. Choose how closely the course follows the source. This is the setting that matters most, and the author picks it.
3. Answer a few questions. The agent asks only what it needs, skips anything it can work out from the material, and checks the answers make sense before it starts building.
4. Watch it build. The planned outline and each phase appear as it goes.
5. Review and refine. The agent stays in the builder once the course exists.

The three fidelity modes

Mode	What it does	When to use it
Verbatim	Keeps the wording exactly as authored	Policies, compliance content, anything already approved
Structured	Keeps the content but reorganises it into a clearer learning flow	Source material that's accurate but poorly sequenced
Adaptive	Rewrites and expands the material into a fuller course	Rough notes, background material, or a thin source

This is the answer to "we don't want AI rewriting our approved content". Verbatim doesn't.

Bring in existing SCORM content

• Convert a SCORM package into an editable course. Upload a package when creating a course with the agent and it becomes a fully editable Coassemble course made of blocks. Text, images, galleries, flashcards, process steps, accordions and knowledge checks carry across, and anything that couldn't be converted is listed for review rather than quietly dropped. Articulate Rise, Articulate Storyline, Adobe Captivate and iSpring are recognised specifically. Other packages are read generically, so less structure survives.
• Host a SCORM package as it is. Upload a package and it runs exactly as authored, with the original design and interactions intact, and attempt-level reporting alongside every other course. A hosted package isn't editable, which makes this the right choice when the original design has to survive.

The customer's original package is untouched either way. Both paths read a copy.

Generate a course from scratch

Describe what's needed in plain language and the agent builds a complete course: structure, screen content and suggested quiz questions. Useful when there's no source document, or when users want a first draft to react to rather than creating from nothing.

Change a course by asking

The agent stays in the builder, so your users describe what they want changed instead of hunting through menus. It can add screens, rework wording, adjust tone or reading level, insert a knowledge check or quiz, reorder screens, split sections and reshape the course.

• Generate screens from a prompt or from existing course content, for filling gaps or expanding a course.
• Generate quizzes drawn from course content. Questions can be reviewed and edited before being added.
• Generate images from a descriptive prompt, added directly to the screen without leaving the builder.
• Rewrite content on any screen, to improve clarity, adjust reading level or shift tone.
• Adjust tone and length, making content more professional, more conversational, more concise or more detailed.

Images

• AI image generation. One generation event produces four options to choose from. Generated images reflect the course topic rather than just the screen they sit on, and generation shows its progress, marking which screen is being illustrated as it works through.
• The built-in stock library is Pexels, searchable from inside the image picker, with images licensed for use in courses.
• Image framing. An author can reposition an image inside its block, dragging to choose which part shows and zooming up to 3x, including inside process, flashcard and slideshow cards.

Review a finished course

• The Review tab. On-demand review of a built course across accessibility, learning design, clarity, engagement and content integrity. Each suggestion names the problem, the screen it's on, why it matters, and the evidence behind it. Most come with a one-click fix, and objective checks are re-run afterwards so a suggestion is only marked resolved once the problem has actually gone. Internally we call this Course Coach, but that name doesn't appear anywhere in the interface, so use Review with customers.
• The Insights tab. The same idea using real learner data, surfacing screens where learners stall, questions most people get wrong first time, and courses taking much longer than advertised. Nothing appears until enough learners have been through the course.

The Review and Insights tabs are available on all current plans, but not on the legacy Flow and Hub plans.

AI narration

• Generate narration from screen text using AI voices, in multiple voices and accents. Narration is embedded into the screen as an audio track.
• Narration drives the screen. When a screen has narration, items can reveal in turn and checklist items tick off as the narration reaches them, so a learner can listen through a screen rather than reading and clicking at once.
• Auto-generated captions. Narration generates closed captions automatically, which can be reviewed and edited before publishing and play in sync during delivery.

Narration is controlled by a toggle in your integration, and that toggle is off by default, so your users only see it once you turn it on.

Translate courses

Translate an entire course into another language in one step, covering all screen text, with the result reviewable before publishing. More than 40 languages are supported, and learners are served the course matching their browser language rather than picking one.

Translation is controlled by a toggle in your integration, and that toggle is off by default, so your users only see it once you turn it on.

Author in your own language

The builder itself is fully localised in eleven languages: English, French, German, Spanish, Italian, Portuguese, Dutch, Danish, Swedish, Norwegian and Finnish. Write to the agent in French and the whole conversation happens in French. This is separate from the language the course is written in, so a user can plan a Spanish course in English if that's easier.

Driving the agent from your own product

Your application can drive the same agent over MCP or REST, so your users get the agent experience without leaving your product. You can scope it to a single course, choose which tools are available, and set it read-only.

Separately, screen embeddables carry an optional AI assistant that's off by default and scoped to editing the screen it's on. That default matters when you're deciding how much AI to expose in your own interface.

A note for partners

Which of these your users actually see depends on the options you enable in your integration. AI features are on by default, and narrations and translations are off by default. Check your configuration before promising a capability to a customer.