Skip to content

Our approach to AI

The principles behind how Coassemble builds and uses AI: what the agent does, what stays under your control, and what happens to your data.

Coassemble uses AI to make course creation faster and more accessible, while keeping you in control of your content and your data. The AI agent does the heavy lifting; you decide what to keep, change, and publish. Here are the principles behind how we build and use it.

You stay in control

The agent acts when you ask it to, and only on your course. It builds and changes content, reviews the course you've made, and reads how learners are going, but nothing is published on your behalf and nothing is generated unless you trigger it.

Everything it produces is ordinary editable content, so you can accept it, change it yourself, ask the agent to change it, or ignore it and build by hand. If your organisation's policies require it, AI can be switched off entirely on Embed.

The agent works alongside you, not instead of you

The agent sits in the course builder as one panel with three tabs:

  • Chat, where you ask for what you want, from a whole course down to a single block.
  • Review, which gives you specific feedback on the course you've built, covering instructional quality, engagement and accessibility, with each finding linked to the screen it refers to.
  • Insights, which reads your course analytics so you can ask where learners are dropping off and act on the answer.

Each of those surfaces a recommendation. Acting on it is your call.

Where you connect an AI assistant of your own over MCP, or drive the agent from your own product, the same principle holds: the tools do what your assistant or application asks for, scoped to the workspace or course you authorised.

Your data stays yours

Your content, prompts, and uploaded documents are never used to train the underlying AI models. You keep full ownership of everything created in Coassemble, whether it was generated with AI or built by hand.

Built on trusted, secure infrastructure

We use established AI models from Google rather than running our own, and we apply the same security controls to AI features as to the rest of the platform: encryption, access controls, SOC 2 Type 1 and Type 2, and GDPR compliance. You can review our controls and reports in the Coassemble Trust Centre.

Honest about what AI can and can't do

AI gets you most of the way to a finished course, not all of it. Like any language model, it can occasionally get something wrong, so we recommend a human review pass before you publish. We build in guardrails to keep generated content aligned with your topic and learning objectives, but you are always the final check on accuracy and tone.

The same applies to the parts of the course AI helps with quietly. Alt text written by the agent is a starting point rather than a finished description, and where it could not produce a meaningful one it leaves the field empty and flags it, so the gap stays visible instead of looking finished.