Voice is not dictation. It's a control surface.
Scan to join —
Dictation turns speech into text — and stops.
The machine heard every word and still did nothing.
Voice accessibility is a productivity unlock, not an accommodation.
Accessibility work and automation work are the same work.
From operating the tool to steering the agent.
When you are thinking three steps ahead, the keyboard is the bottleneck.
Say it. This deck should advance right now.
FALLBACK — if voice does not fire, press Down yourself and narrate: "the key is the same one your voice was pressing."
Same words, from my phone. One command, two entry points.
FALLBACK — one command, two entry points. If the phone path is down, say so plainly and press Down.
Say the task. Watch it land in the tracker.
FALLBACK — show the task already created, and narrate the creation you did live.
Ask what is running. Hear the state read back.
FALLBACK — the task from the previous slide is already on screen; read it aloud instead of re-querying.
Speak. The agent does the typing.
FALLBACK — recorded clip. Decide this before rehearsal day, not during.
Scan the code in the corner. It is a throwaway sandbox — break it.
FALLBACK — prerecorded clip. Do not attempt live audience input unless it has been tested safe.
Parlay's API has no authentication today.
So it stays on the private network.
Public exposure is a deliberate decision, not a default.
An experience, not a link list: first useful commands, then voice-first development, then agent orchestration.
Try the first challenge tonight — it runs in your browser.
Not the fallback. Not the gimmick.
Thank you.