How `riff` shows up here In this channel's material, riff describes the exploratory, talk-it-through style of working that the live voice experiment is built around. The whole session is framed as a working experiment, not a finished feature — Kieran thinks out loud in real time, reacting to a website by pointing, drawing, annotating, and simply speaking, so an agent can start working with what he's noticing right away. [^edd078ec-d05b-519d-bd92-9769eae6f9a1@167] [^edd078ec-d05b-519d-bd92-9769eae6f9a1@261] Why riffs matter for live voice The riff is the quality signal the experiment is chasing: the spontaneous spoken reactions, questions, and half-formed instincts a person produces while looking at real work. Capturing those moments — rather than a polished written brief — is exactly what OpenAI's live audio API plus screen capture makes possible in this setup. [^edd078ec-d05b-519d-bd92-9769eae6f9a1@213] It also shapes the working rhythm: AI tends to be conservative, so a riff that starts from a big-picture frame (what would I do on a year-long project?) and then zooms in is how abstractions surface. [^edd078ec-d05b-519d-bd92-9769eae6f9a1@1095] - riff = exploring an idea by talking through it freely, in real time, rather than writing a polished spec.- In this video it's the mode of work the live voice experiment enables — spoken reactions, questions, and instincts captured as the quality signal.- Riffing works best in the polish phase, where you can see the result and just say what to change. [^edd078ec-d05b-519d-bd92-9769eae6f9a1@804]