Say it instead of typing it

The assistant asks you to describe your career, and a keyboard is the worst possible instrument for that. Someone who would happily talk for two minutes about a project types two sentences instead — and two sentences is what ends up in the experience bank.

There is now a microphone in the composer. Press it, say what you mean, and the words land in the message box. They are not sent for you: a transcription is a guess, and a guess sent on your behalf is a message you cannot unsend. Read it, fix it, send it — or discard the recording with Escape and nothing leaves your browser at all.

Two smaller things landed with it.

The reply suggests what to ask next. After the assistant answers, up to three questions appear underneath it — compare the first two roles, tailor my CV to this one, which of these pays the most. Clicking one sends it as if you had typed it. They only show up after an answer that went through, never after something went wrong.

Scrolling up during a long answer now works. The transcript used to yank itself back to the newest line every time another word arrived, which made re-reading what was said a minute ago impossible on exactly the answers worth re-reading. It now follows along only while you are already at the bottom; scroll up and it stays put, with a Jump to latest button to bring you back.

The assistant lives at /my/assistant, and the microphone appears wherever the composer does — including the CV tailoring workspace and the experience interview.