Latent Space panel calls OpenAI's DevDay apps "regressions": Dots hidden in UI, Ultrafast eats a weekly reset in 20 minutes

#What happened

A Latent Space podcast roundtable of AI practitioners — Dex Horthy (founder of Raindrop), Ben Hylak (founder of Human Layer), Alex Volkov, Ben Davis, and Allison Durham — delivered a blunt verdict on OpenAI's DevDay releases: negative on product, mildly positive on one model.

#Per-release verdicts

Release	Panel verdict
Dots	"right idea, wrong packaging" — deeply hidden within the UI; Hylak says it needs to be a standalone app
GPT-6.1 Sol	underrated, good price — but branding confusion after GPT-6 Sol shipped a week earlier
Ultrafast mode	useful but expensive — Ben Davis burned a full weekly reset in about 20 minutes
Codex desktop app	"regressions" — extra sidebar, worse performance; Horthy's input field freezes when the machine is offline
Decision models / Jev	Hylak's Jev-based agentic memory pipeline cut cost 10 to 20 times; Davis calls viral Jev demos "aggressively stupid" — his favorite real use case is command-K menus

#Key quotes

• Dex Horthy on over-helpful assistants: "Prompt injection is definitely not the only thing I'm worried about... There's a lot of other ways to cause your life a lot of harm that is not other people trying to do it. The model is actually trying to be helpful. Stupid is much more dangerous." He cited an incident where someone asked a friend's AI assistant for a Netflix two-factor authentication code from their email — and got it.
• Horthy trusts exactly three companies with that data: Apple, Google, OpenAI ("There's not a fourth for me"), explicitly excluding Amazon and Facebook.
• Horthy on the Codex desktop app's shifting behavior: "things randomly change all the time... you squint, and you're like, wait, what?" OpenAI renamed "fork chat" to "copy chat" because non-developers wouldn't understand "fork."
• Grokbot anecdote: roughly 1 in 5 AI-drafted emails contains details he does not want sent; one message once went out signed "this person's Grokman."
• The panel's prediction: users will converge on a single personal assistant because of the attention cost of switching tools.

#Why it matters

Even sympathetic, working AI practitioners are now rating OpenAI's product execution negatively while nodding at the models — a sign the trust and UX battle, not raw model quality, decides who owns the personal-assistant slot everyone expects to consolidate.

Verified October 17, 2026 — story confirmed against a source page published 2026-10-17T01:08:28Z.