.png)
If you've seen AI from Cadmium so far, you've seen one thing: the AI Writing Assistant inside Elevate. It's useful, but modest, and it's fair to wonder whether that's the whole plan.
It isn't. That feature is the first small step we took in public. The larger work has been happening underneath it, at the platform level, where customers can't see it yet. This article is about that foundation, and why bringing AI to associations and continuing education the right way starts there.
One investment, every product
The easy path would have been to bolt a separate AI feature onto each product, one at a time. That gets something on a screen quickly, but it also means five different implementations, five quality bars, and five things to maintain and secure.
We went the other way and built shared AI infrastructure once, so every product in the integrated suite can draw on it. The payoff is simple:
- One investment instead of five.
- Consistent quality and consistent guardrails across every product.
- Faster AI in each new product, because the hard part is already done.
Three parts do the work under the hood, and each earns its place by what it does for you:
- A single control panel where our teams set up and adjust the AI without waiting on an engineering release.
- An engine that runs it securely, on Amazon's managed AI, inside a proven enterprise environment rather than something we stood up on our own.
- A drop-in that places AI directly inside the tools your team already uses, so the help shows up where the work happens and there's nothing new to learn.
The first feature was small on purpose
The AI Writing Assistant went live in Elevate on April 2026, and it's in daily use now. It takes the friction out of the writing that piles up around a learning program:
- Draft a course description from a short prompt instead of a blank field.
- Tighten loose wording before it goes out.
- Shift tone between professional, casual, executive, and empathetic.
- Make text shorter or expand a rough note into something readable.
- Pull a clean summary out of a dense block of copy.
The modest scope was the point. If a draft isn't quite right, the admin edits it, and nothing is at stake beyond a few keystrokes. That let us put a real tool in customers' hands while we strengthened the foundation underneath, so the more ambitious features that follow arrive on something proven rather than something rushed.
Changes are tested, versioned, and reversible
A fair question about any AI tool is what happens when it changes, because these systems do change. Ours are set up so that change is controlled:
- Our teams can adjust how the AI behaves without a software release.
- Every change is saved as its own version, so previous behavior is never lost and a rollback takes seconds.
- Changes are tested in a safe area first, away from live accounts.
- A separate AI reviewer scores responses against clear standards, including relevance, accuracy, and safety, and we compare each version against the last before anything goes live.
If the numbers don't move in the right direction, the change doesn't ship. When we implement any changes, that's a decision we made, tested, and can account for.
What this means for regulated programs
In a regulated field, the provider stays responsible for what learners see, and AI doesn't change that. Our aim is to provide the controls, the grounding, and the records that support the work you're already accountable for. A few of those controls are live today:
- AI can be switched on or off at the account level, so it stays out of programs where you'd rather it not touch anything yet.
- Your team defines how the AI behaves, and every change to those rules is saved with a full history to review.
- The AI only surfaces content a given user is allowed to see, and it doesn't send personal information to the AI models.
The next item on our roadmap is to ground the AI in a customer's own approved content and instructions, so answers reflect their materials and accreditation language rather than generic model knowledge. Compliance work stays with the customer, along with the controls and the evidence to support it.
What's coming next
The intelligent search understands what a learner means in plain language. Picture typing "the CE courses I can finish this week" and getting real, permitted results back, each with a plain reason it matched. The first version focuses on finding the right content, not on generating answers, because in this market a confident wrong answer is worse than no answer.
To see what it does inside your own Cadmium products, talk to your account team or get in touch with an expert here.
