TL;DR
The short version
Claire Vo separates DevDay into three bets: Dots (good work quality, muddy information architecture), Space and Sites (what enterprises asked for on human+agent documents), and Decisions API plus Astra Ultrafast (product shapes latency used to forbid — at a real burn rate).
From her How I AI walkthrough: she still prefers micro-agents over one Dot; Ultrafast demos cost about $97 for half an hour of play.
Claire Vo, How I AI, recorded at OpenAI DevDay, published 30 Sep 2026.
Dots: good bones, wrong map
A Dot is always-on across ChatGPT, Codex, computer, phone, call, and Slack — one issued agent for now. Early shopping, coding, and proactive schedule help impressed. Thread taxonomy did not. Claire stays on micro-agents hired per job and waits for a full Dot review until the surface stops moving hourly.
The bones of your Dot are good. There are a bunch of rough edges around the user experience.
Claire Vo
Space and Sites: collaboration, not chat
Space is the underhyped enterprise feature — docs, slides, HTML artifacts shared by humans and agents with controls. Sites bundle connectors so vibecoded dashboards honor each viewer's permissions. That answers the "what do we do with all these prototypes" problem she hears from companies. Google still owns the default; this is the shot.
Decisions with eyes; Ultrafast with a bill
Sol 6.1 is the workhorse price cut. Decisions API is Jev-shaped with vision — constrained choices for product plumbing. Astra Ultrafast (~8× speed / ~6× cost) enables live SVG and realtime 3D edits. Her kids played for thirty minutes; the bill was about $97.
