Week 6 Lab: Inbox smart-reply audit

A lab auditing your own email or messaging app's suggested replies against the messages that triggered them

1 April 2027

Before the lab#

Bring a device with an email or messaging app that offers suggested replies, and the ability to screenshot or otherwise record what it suggests. You are auditing your own inbox, not a shared example set, because the suggestions are personalised to you specifically.

In the lab#

Scroll back through recent messages, screenshotting the suggested-reply row for each one before you act on it. Ten is the minimum; the more varied the messages, the more useful the audit — mix routine logistics with anything emotionally loaded that came through the same channel.

Group discussion focuses on the near-misses rather than the outright wrong ones, since those are more revealing: a suggestion that’s almost right, but slightly too cheerful or slightly too brief for what it’s replying to, shows the model’s actual granularity better than a clean failure does.

Afterwards#

If nothing in your ten messages was wrong, that’s a real result, not a failed lab — say so and say why you think your inbox didn’t trigger a failure, which is itself evidence about how these systems are scoped.