A guided demo, clearer AI drafts, and a hardened staff assistant connection

This week's thread is how First Six is shown and how its AI features are held to account: a demo that guides instead of overwhelming, a clearer warning on AI drafts, and more protection around an assistant connection that is not yet available to any university.

A demo that starts from what you came to see

Anyone opening the demo now begins by choosing a goal, such as "Students leaving before census" or "Help-seeking and crisis safety". A small panel then walks them through three to five real screens for that goal. For each one it says what the screen shows and why it matters, outlines the part of the page it means, and follows the visitor between the student app and the console. It also answers about twenty common questions in First Six's own words.

The walkthrough is scripted. It does not call an AI model, it stores nothing except a short note in the visitor's own browser, and it sends nothing anywhere.

On a branded preview (the links we send to individual institutions), the walkthrough only points and explains. It does not invite changes to the shared personas, and it skips the one screen that lists a real person's name.

What did not change: the screens the walkthrough shows are the ordinary product screens, and nothing a visitor does in a demo reaches a real institution's data.

A calmer demo, and shared previews that stay unchanged

The demo used to open under the dimming product tour, the go-live checklist, the "First time here?" notes and the student onboarding cards. Those are meant for a new customer's first week, so they no longer appear in a demo. Real institutions are unchanged.

A branded preview now says "Live preview" rather than counting down, and what one visitor does no longer waits for the next. Tour dismissals, onboarding progress and guide progress are no longer saved against the personas that every visitor of a preview link shares. The demo's ended pages and sign-in screens also stop quoting a fixed length of time or claiming the data was already cleared, which was not true in every case.

The AI tools in the website demo are now metered, so a demo cannot run up unbounded use.

Write with AI shows that drafts can be wrong

Write with AI now shows, on screen beside the draft, that its drafts can be wrong and need checking before they are published. This matches the wording the other AI features already carry. Nothing goes live until a person publishes it.

More protection around the optional staff assistant connection

As described last week, the optional connection that would let a staff member use their own AI assistant with the console is not available to any university and is off unless an institution chooses to switch it on. It is currently enabled only for First Six's own test tenants, none of which hold real student data. Two further security reviews this week led to these changes:

• Insights shown to an assistant are now protected against small groups by one consistent set of rules. Nothing is sent as an exact count, and anything withheld takes the whole report with it. Leaving reasons appear only when at least five different students chose them, so one student's reason cannot be picked out.
• Every sign-in token is now tied to its own connection, so renewing a connection cannot reset its limits and a leaked old token cannot hold a live connection at its limit.

What did not change: the assistant can still only prepare drafts for a person to review, never publish, and it receives no student record or anything that identifies a student. No customer institution has been switched on.