Stratechery: the direction behind OpenAI's confusing Dev Day; Lenny's: the releases that matter, and Jev's 8 real use cases

AI platforms: OpenAI's Dev Day, argued and field-tested

"OpenAI Dev Day, Dot and OpenAI's Product Transition, Sign In With ChatGPT" — Stratechery

1. Thompson's summary line is the public verdict in full: "OpenAI's Dev Day showcased a product that is, frankly, pretty confusing. However, there is more vision here than it might seem." The Update lands the morning after an event that ran Tuesday, and its three markers sit in the same territory as Thompson's free essay of September 28, "Apps, Agents, and Aggregation," which this digest carried on September 29 — the argument that agents become the aggregators and apps shrink into implementation details. 3
2. "OpenAI Dev Day" and "Dot" carry the event and its headline release. OpenAI held Dev Day on September 29 in San Francisco, and its own recap counts more than 20 announcements across ChatGPT, Codex, its models and new ways of working, for a product it says reaches 1.2 billion weekly users. 4 Dots, the release the marker names, are always-on agents powered by GPT-6 Astra, each with its own cloud computer and browser, rolling out across Pro, Business Premium and Enterprise plans in eligible markets. 5
3. "Sign In With ChatGPT" is the half of the transition most likely to touch a product team's roadmap. Other products can now sign users in with a ChatGPT account, and eligible Plus and Pro subscribers can spend their ChatGPT plan allowance inside those tools; the first cohort is 16 partners, among them Cognition's Devin, Notion, Vercel, T3, OpenClaw and Dactyl. 4 Identity is available globally, while the shared-allowance half covers Plus and Pro users in the participating tools.

"OpenAI Dev Day 2026: The releases that actually matter" — Lenny's Newsletter (How I AI)

1. Her first item is Dots, and her verdict there is provisional. Vo walks through what Dots can do and how she has been using her own, and holds the final judgment until she has lived with one longer; the early impressions, rough edges included, come first in the chapter list. 6
2. Spaces and Sites are the releases she expects people to underrate. She calls Spaces one of the most underhyped announcements of the day for what it does to collaboration between people and agents, and shows Sites with connectors and plugins as the way a team shares an internal tool while each teammate keeps their own data permissions. 6
3. The model-and-API half is where she gets hands-on. She places GPT-6.1 Sol in her own stack on speed and cost, uses the Decisions API with vision to pick podcast thumbnails, and builds a collaborative sketchpad and a promptable 3D world on Astra ultrafast; that last experiment set her back about $97. 6

Building with AI: what a model that only decides is good for

"Jev: 8 real use cases for the fastest, cheapest model I've ever used | John Lindquist" — Lenny's Newsletter (How I AI)

1. Jev answers in a fixed vocabulary — a category, a score, a probability — and every demo is built on that shape. TypeSafe AI's model returns one of a set of predefined values, which is what lets it sit inside a live interaction. Lindquist opens on what Jev actually outputs, then on the sequential calls that chain from it. 7
2. Four of the demos are real-time or high-volume work. A voice to-do app that classifies and executes spoken commands with no visible pause; a deduplication pattern that merges messy records in milliseconds using confidence scores; Jev as a router that takes one line of plain English and walks a user deep into an app; and a chess match against a low-reasoning model that puts speed and cost side by side. 7
3. The closing stretch is the boundary. The episode ends on where Jev falls short and when a full generative model is still the right tool, plus the multi-step classification pattern Lindquist reaches for when one Jev pass is not enough. 7

Where to start