Dan Shipper, CEO of Every, breaks down the main OpenAI DevDay 2026 announcements after early access and a week of hands-on testing: Dots, an always-on agent; Space, a shared home for docs, slides, sheets, and files; plugin extensions and discovery; Sign in with ChatGPT; and the Decisions API. He also covers Astra Ultrafast, Sol 6.1, updates to the Agents API and Codex CLI, and the new $500 plan, arguing that persistent agents feel like the next form factor for AI while advising viewers to wait before relying on this one. Chapters: 00:00 Intro 00:29 Dots, the always-on agent 02:38 Space for docs, slides, and sheets 04:01 Plugin extensions and discovery 05:07 Sign in with ChatGPT 06:10 Decisions API 07:09 Astra Ultrafast, Sol 6.1, Agents API, and Codex CLI ## Channel Identity Observations ### Self-Description Evidence - **Runs hands-on product tests.** The title states "We Tested OpenAI DevDay Products!" and the creator says: "I've had a chance to test this out over the last week or so" (regarding Dots), plus "During our internal testing at Every, we found it to be faster than Jev for some workloads" — indicating the channel is tied to the organization **Every**. ### Intended Audience - **AI practitioners and developers.** Direct address: "So, if you're a developer and you're interested, you should definitely check out the solutions API." Discussion of pricing plans, tokens/second, and APIs assumes technical literacy. - **Builders/startup-oriented viewers.** Commentary like "a new distribution mechanism for developers and startups" and "People like us should always be careful about how a user enters our app" positions the audience as people building AI products. ### Claimed Scope - **Early-access AI product coverage.** "We got access to everything they had released before" — the channel claims pre-launch access and positions itself as a fast, first-hand breakdown ("what they launched, what's important, and what it means for you"). - **Historical context framing.** The creator cites four years of DevDay attendance ("This is my fourth DevDay") and traces form-factor eras from "GPT-3 era" to "the era of permanent agents," claiming industry-level analysis authority. - **Broad scope across user types:** "It's a mix of things for knowledge workers and consumers, and a bunch of things for developers." ### Demonstrated Presentation Patterns - **Numbered listicle format** (items 1–6: Dots, Spaces, plugin extensions, Login with ChatGPT, Solutions API, other launches) with balanced pros/cons — e.g., praise for Dots ("clearly the future") alongside critique ("I found lots and lots of bugs"). - **Actionable recommendations** ("my recommendation is to wait a week or two and then try"). - **Cross-promotion of upcoming content:** "I'll be interviewing Sam Altman on this channel" and "We'll also be releasing a DevDay vlog" — evidence of interview access and a multi-format content strategy. - **Insider industry perspective:** familiarity with competitors (Muse, Grok bots, Openclaw) and internal figures like Pete Steinberger, presented as a well-connected industry commentator rather than a neutral reporter. ### Note on Source Material Some transcript terms (e.g., "Dots," "Space," "Astra Ultrafast," "GPT-6 Luna") appear only here and could not be cross-verified; observations are limited to what this video itself demonstrates. # Delivery Observation: DevDay Products Breakdown ## Overall Format - **Numbered listicle structure.** Each launch is introduced with an explicit ordinal: "Number one is dots…", "Okay, number two, space", "number three", "Number four", "Okay, number five", "Sixth". This is the video's dominant organizing device. - **Live-event framing.** The speaker anchors the delivery to the moment: "We are here. We're in San Francisco", "If you're watching this, it means the keynote has just started." - **Fast pacing with short segments.** Six items are covered in roughly nine minutes, with the last item bundling multiple launches ("a bunch of other things"). ## Recurring Evaluation Habits - **"It's X, but not quite ready yet" pattern.** Dots is described as: "When it works, it's clearly the future… But it just works. Sometimes." And later: "from my testing, it's not quite ready yet. It's just that all these little things don't quite work to make this experience great." - **"I've been asking/wanting this for a long time" as endorsement.** Used for Spaces ("I've been wanting them to do this for a long time, and it finally happened") and plugin extensions ("This is something I've been asking for for a long time, and it's finally here"). - **Direct recommendations to the viewer.** "you should definitely check out the solutions API", "you should definitely check out the Soul 6.1", "You should check it out", "my recommendation is to wait a week or two and then try." - **Self-correction mid-sentence.** "But it just...It just works. Sometimes. W-when it works, it's good." Also a spontaneous "What is number three? Oh, that's good." — suggesting lightly scripted delivery with real-time reference checking. ## Vocabulary and Phrasing - **"Point" as a personal noun.** The speaker refers to their dots agent as "my point": "My point is called Boo", "I can mark my point, and I can come in and work with me". - **Intensity adverbs.** "truly insanely fast", "incredibly fast", "This is crazy", "I think it's incredible", "really, really fast". - **Pragmatic comparisons to everyday systems.** "your AI subscription will be very similar to a mobile phone plan", "a Microsoft Office suite inside ChatGPT", "almost like Google Drive." - **Casual closers.** The video ends with: "And see you in space. I don't know." — an offhand joke left in. ## Teaching Habits - **Quantified framing.** Historical launch counts open the video: "they launched 11 products… Second year–five. Third year–nine. This year, 22." Concrete specs recur: "150 milliseconds", "250 tokens per second", "50% of the price", "about 15 apps." - **Personal testing as evidence.** "I've had a chance to test this out over the last week or so", "During our internal testing at Every, we found it to be faster than Jev for some workloads." - **Historical arc framing at the end.** "Every 6-12 months you see how the model progresses creating a change in form factor… And now we are entering the era of permanent agents." ## Isolated Moments (Low Repetition Confidence) - The **interview teaser** ("I'll be interviewing Sam Altman on this channel") appears once and cannot be confirmed as a recurring habit. - The **"operating system" thesis** ("They are creating an operating system to run all your work inside ChatGPT") is stated once; whether it's a signature framing across videos is uncertain from this transcript alone. - The **false starts and "I don't know" ending** could indicate a live, unedited recording style, but a single transcript cannot confirm this is consistent. ## Uncertainty Notes - Transcript quality is uneven (e.g., "Openclaw", "Grock bots", "point" for what is likely a differently named product), so quoted product names reflect the transcript as given. - The speaker's identity, channel branding, and whether the numbered format is habitual are not verifiable from this single video. # Video Observations: OpenAI DevDay Product Breakdown ## Source & Speaker - **Video:** "We Tested OpenAI DevDay Products! 6 Things to Know: Dots, Spaces, Astra Ultrafast" - **Speaker:** The narrator is identifiable only as the video's host speaking in first person; no name is supplied in the transcript or metadata. He states he is at DevDay in San Francisco and this is "my fourth DevDay." - **Evidence basis:** All observations below come solely from the transcript; no private beliefs are inferred. --- ## Claim 1: DevDay Scale — 22 New Products/Features - The host claims OpenAI launched **22 products and features this year**, versus **11 in year one, 5 in year two, and 9 in year three**. - Presented as context-setting, framed as an escalation in release volume. --- ## Claim 2: "Dots" — Persistent Agent **Position:** Dots is "clearly the future" when it works, but is not yet ready. **Stated reasoning:** - Described as a "constantly active persistent agent," comparable to Openclaw, Muse, Instinct, or Grokbots. - Frames it as the conclusion of a story arc: Openclaw "completely failed," and OpenAI hired Pete Steinberger (Openclaw's creator) about nine months ago to build persistent agents. - Dots appears as a new tab in ChatGPT; it is one continuous chat, has access to its own computer and the user's computer, and is proactive (texts the user unprompted, e.g., answering incoming planning requests). **Positive assessment (after ~a week of testing):** - "When it works, it's clearly the future." - He reports starting to use it more than starting new chats in Codex or ChatGPT. **Qualifications (prominent):** - "Sometimes. When it works, it's good." - He found "lots and lots of bugs and inconsistencies"; Dot itself struggled to interpret ambiguous requests (e.g., which "browser" the user means). - Problems with logging in, accessing external sites, and using his wallet. - Explicit recommendation: **wait a week or two before trying it** — "it's not quite ready yet." --- ## Claim 3: "Space" — Office Suite Inside ChatGPT **Position:** Good implementation with a structural advantage, but adoption is an open question. **Stated reasoning:** - Described as "OpenAI's attempt to create a Microsoft Office suite inside ChatGPT" — file system (like Google Drive), document editor, slides, and spreadsheets, all native. - Enables shared parallel workspaces between humans and agents (e.g., leaving a comment, tagging his "point," collaborating). - Key advantage: users "never have to leave ChatGPT." **Qualification:** He explicitly withholds judgment on whether it will replace Google Slides/Sheets/Keynote/Notion: "It's a difficult task," and he wants to see whether they become "fully functional enough" to move over. --- ## Claim 4: Plugin Extensions **Position:** The most promising of OpenAI's repeated attempts at an app store. **Stated reasoning:** - Described as native Codex applications that developers can build and distribute to companies/users, usable alongside Dots or regular chats. - No plugin store; instead ChatGPT automatically discovers and surfaces plugins when users ask for something — presented as a **new distribution mechanism** for developers and startups. **Qualifications:** - "It remains to be seen how good it is." - Notes OpenAI has had "three or four cracks at creating an app store," with this appearing, from demos, to be the most promising attempt. --- ## Claim 5: Login with ChatGPT Expansion **Position:** "This is incredible" — good for OpenAI, developers, and users. **Stated reasoning:** - Users can spend their ChatGPT subscription tokens in ~15 third-party apps (example given: Devin), reducing usage costs there. - For app developers (like himself, "people like us"): it removes the incentive to use cheaper models, since developers normally worry "how much AI are they using?" — with bring-your-own-tokens, developers "can use the model that works best for you." - For users: AI subscriptions become "very similar to a mobile phone plan" — tokens usable anywhere. **Qualification:** He calls it "the first sample of this." --- ## Claim 6: Solutions API **Position:** Developers interested in fast structured answers should check it out. **Stated reasoning:** - Presented as OpenAI's answer to a model called "Jev" (recently released), which gives fast structured answers usable as a classifier/estimator, at a much lower cost (~"a couple hundred times cheaper" than Fable-level performance for many tasks). - The Solutions API is based on a fine-tuned version of **GPT-6 Luna**, gives answers in ~150 milliseconds, is multimodal (accepts images), and in the host's company (Every) internal testing was **faster than Jev for some workloads**. --- ## Claim 7: Additional Launches - **$500 plan:** more usage than the $200 Pro plan; includes access to **Astra Ultra Fast**, a new speed tier processing **250 tokens per second** ("incredibly fast"), discounted at launch. - **Qualification:** the speed "will deplete your token balance very quickly, so be careful." - **Soul 6.1:** Astra-level performance at 50% of the price; recommended for those who haven't paid for Astra. - **Agent API update:** enables building apps at Codex's level, now including computer usage. - **New Codex CLI:** voice chat, improved agent-tracking interfaces. --- ## Overarching Thesis - OpenAI is "creating an operating system to run all your work inside ChatGPT" plus a **platform/ecosystem** for developers to build inside ChatGPT and carry subscriptions across the web. - **Form-factor cycle claim:** every 6–12 months, model progress drives a form-factor shift — GPT-3 era: ChatGPT on the web; Sonnet 3.5: terminal coding agents; ~February: GPT-5.3 into desktop apps; now: **the era of permanent agents**, of which Dots is "OpenAI's first step." - Final stance: promising form factor, strong positioning via ChatGPT integration, but "from my testing, it's not quite ready yet" — recommendation to wait one to two weeks. --- ## Unresolved Tensions & Notes - The host is simultaneously enthusiastic ("clearly the future," "incredible") and critical ("I just didn't have the best experience with this product") — both stances are preserved as stated, not reconciled. - Several assessments are explicitly provisional: Space's adoption, plugin quality, and Dots readiness are all framed as "to be seen." - Product names in the transcript (e.g., "Jev," "GPT-6 Luna," "Soul 6.1," "Astra," "Openclaw") are recorded as expressed in the source material. - Upcoming content teased: interview with Sam Altman and a DevDay vlog on the same channel. # Video Attribution Record ## Video **Title:** *We Tested OpenAI DevDay Products! 6 Things to Know: Dots, Spaces, Astra Ultrafast* **Video ID:** 01a0f15a-4412-7fa4-a0b1-bffc828ad386 ## Speakers & Roles | Speaker | Role | Evidence | |---|---|---| | Unnamed host | **Creator / presenter** | First-person narration throughout: "We are here… This is my fourth DevDay"; hands-on product testing ("I've had a chance to test this out over the last week or so") | | Company affiliation | **Every** (stated) | "During our internal testing at Every, we found it to be faster than Jev" | | Sam Altman | **Upcoming guest (announced, not present)** | "I'll be interviewing Sam Altman on this channel" — interview has not occurred in this video | No guest, narrator, or third-party speaker appears in the transcript. Single-speaker format. ## Referenced Sources & Third-Party Claims - **OpenAI DevDay** — primary subject; host reports 22 products/features launched this year, with prior-year counts (11, 5, 9) per the host's recollection. - **Named third-party products/claims (host's characterizations, unverified):** Openclaw, Muse, Instinct, Grokbots, Devin, Jev, Fable, GPT-6 Luna, Sonnet 3.5, GPT-5.3, Notion, Google Slides/Sheets. - **Pete Steinberger** — claimed creator of Openclaw, reportedly hired by OpenAI per the host's narrative. - **Performance/price claims** (host-reported, from demos or personal testing): Solutions API ~150 ms responses, "couple hundred times cheaper" than Jev; Astra Ultra Fast at 250 tokens/sec; Soul 6.1 at "50% of the price" with "Astra-level performance"; ~15 apps in login-with-ChatGPT program; $500 plan vs $200 Pro plan. - No cited documents, links, or external reports are given; all claims are from the host's testing, demos, or recollection. ## Promotional / Self-Promotional Segments - **Upcoming content promotion:** Interview with Sam Altman "tomorrow" on this channel; a forthcoming "DevDay vlog." - **Affiliation disclosure:** Host speaks as part of **Every** and mentions its internal testing; also references "people like us should always be careful about how a user enters our app," indicating the creator operates an app business. - No paid-product sponsorship is identifiable from the transcript. ## Availability & Access Limits - **Curated metadata:** None supplied beyond title and transcript. - **Comments:** No information available. - **Access note:** Host states "We got access to everything they had released before" — early/pre-launch access to OpenAI products. ## Uncertainty Flags - Speaker name is never stated in transcript or metadata; creator identity is inferred only from the "Every" affiliation mention. - All model/product names (e.g., "Dots," "Space," "Jev," "Astra") appear only via this transcript and could not be cross-verified. We are here. We're in San Francisco, and we're going to talk about DevDay. This is my fourth DevDay. In their first year, they launched 11 products and features. Second year–five. Third year–nine. This year, 22 new products and features are being released today at DevDay. If you're watching this, it means the keynote has just started. We got access to everything they had released before. Here's our breakdown of what they launched, what's important, and what it means for you. Number one is dots, a constantly active persistent agent that works much like Openclaw, Muse, Instinct, or Grokbots. This is the conclusion of a story arc that began about nine months ago, when Openclaw completely failed, and OpenAI hired Pete Steinberger, who created Openclaw, to help them create such persistent agents. So , this is OpenAI's entry into this race. Dots is a constantly active permanent agent. It appears as a small new tab at the top of ChatGPT. So, instead of starting a new chat in ChatGPT for Work or Codex, whether in the desktop or mobile app, all you have to do is click on your point. My point is called Boo. Then you have one long chat that you can use for any action in AI. Your resident agent has access to their own computer, they have access to your computer, and they are proactive. So you'll notice that every now and then throughout the day he'll just text you, "Hey, I did this. Hey, for example, you've had a few people contact you to plan something with you . I answered them." Something like that. When it works, it's clearly the future. This is really great. I've had a chance to test this out over the last week or so, and I'm starting to turn to it more than I would to starting a new chat in Codex or ChatGPT. This is one of those things where you think, "I could have done this sooner if I had manually set up all this with heartbeats and such." But it just...It just works. Sometimes. W-when it works, it's good. The only problem is that, at least in the versions I tested, this isn't quite the case yet. If you compare this to Muse or Grock bots, it's just a very early product that they're developing very quickly . And I found lots and lots of bugs and inconsistencies that made it a little confusing for me to use, and honestly, Dot herself had a hard time figuring out what to do. If you ask it to open a browser in the app, it doesn't know whether you mean a browser in its own thread, a separate Codex thread, or on its cloud computer. I had many problems, such as logging in, accessing external sites, and using my own wallet. And all these things are getting fixed, and I...I imagine they'll be different in a week or two. But you can feel the speed at which OpenAI works, and I just didn't have the best experience with this product. Okay, number two, space. Space. Space. Space is OpenAI's attempt to create a Microsoft Office suite inside ChatGPT. I've been wanting them to do this for a long time, and it finally happened. It's a native implementation of something like a file system, almost like Google Drive, plus a document editor, slides, and spreadsheets, and you can do all of this from your ChatGPT or Codex task, or if you're using your point. What's really cool is that it's all in one place. So, I can leave a comment in the document, and then I can mark my point, and I can come in and work with me, and it's kind of like a parallel shared space. Space for people, and now for agents. And you can see how if I use this with my point and other people in my company, how I never have to leave ChatGPT, which I like. I think the implementation of the space is quite good. So, I'll be interested to see today and over the next few weeks whether they'll be fully functional enough for me to move on to using them fully. Do you know if I access them in Google Slides, Sheets, Keynote, or Notion? It's a difficult task, but I think for the space and for the dots, the integration into ChatGPT itself is a huge advantage . Okay, number three. What is number three? Oh, that's good. They create something called plugin extensions. Plugin extensions are essentially similar to native Codex applications. So, just as they run their own native versions of Docs, Slides, and Sheets, plugin extensions allow you to create apps for yourself that you distribute to your company or your users. These are applications designed to be used in ChatGPT in parallel with dots or simply with regular ChatGPT or Codex streams. This is something I've been asking for for a long time, and it's finally here. And they added the option to let ChatGPT automatically find your plugin. So they don't have a plugin store, and they've created a way where when users ask ChatGPT to do something, it will look for and potentially bring up your plugin, your app, which the user will use as a new distribution mechanism for developers and startups. It remains to be seen how good it is. They've had three or four cracks at creating an app store, but from what I've seen, from the demos I've seen, this is the most promising option right now. Number four. This is a good option. They have expanded the login with ChatGPT, so you can take your ChatGPT subscription and spend your tokens on third-party apps. They have about 15 apps that are participating in this right now, so apps like Devin where you pay for a ChatGPT subscription, when you switch to using Devin, instead of using your usage on Devin, you can take your ChatGPT subscription with you and reduce your usage there. I think it's incredible. This is really good for OpenAI . This is really good for app developers . People like us should always be careful about how a user enters our app. We need to be careful about “how much AI are they using?” ". This incentivizes app developers to use cheaper and cheaper models, and all that kind of stuff . Now, if you bring your tokens, we don't have to worry about that. You can use the model that works best for you. And that means that, in my opinion, this is a really good deal for users. This means that your AI subscription will be very similar to a mobile phone plan. And you can use your tokens anywhere. This is the first...This is the first sample of this, but I'm really happy with it. Okay, number five, we have a solutions API. If you've been following anything in AI, you know that a few weeks ago this new model came out called Jev, which gives you really, really fast, uh, structured answers to questions that you can use as an LLM , as an estimator or a classifier. It's not quite an LLM, it's something like a slightly different model, and it's much cheaper. It gives Fable-level performance for many tasks, but I don't know, it's about a couple hundred times cheaper. And OpenAI is releasing it today. It is based on a finely tuned version of GPT-6 Luna, and it is truly insanely fast. Uh, you can get answers in about 150 milliseconds. During our internal testing at Every, we found it to be faster than Jev for some workloads. And, most importantly, it's multimodal, so you can send images to it. So, if you're a developer and you're interested, you should definitely check out the solutions API. Sixth, they launched a bunch of other things. Good. They are launching a $500 plan. It has a lot more usage than the regular $200 Pro plan and gives you access to Astra Ultra Fast, their new Astra speed tier, which is also discounted today. It processes 250 tokens per second. It really is incredibly fast. This is crazy, but it will deplete your token balance very quickly, so be careful. We also have Soul 6.1, so they are launching a new version of Soul. It has Astra-level performance, but costs 50% of the price. If you like the Astra but have n't paid for it, you should definitely check out the Soul 6.1. They are also releasing an update to their agent API that allows you to build your own applications at the same level as Codex, and it now includes computer usage. Last but not least, they are releasing a brand new Codex command line interface. It has voice chat, new and improved interfaces for tracking agents, and a bunch of other things. You should check it out. Okay, that's all they launched at DevDay. You can see that it's a mix of things for knowledge workers and consumers, and a bunch of things for developers. They are creating an operating system to run all your work inside ChatGPT, and they are creating a platform and ecosystem for developers to build applications inside ChatGPT, and to bring your ChatGPT subscription to other places on the web. Every 6-12 months you see how the model progresses creating a change in form factor. So, in the GPT -3 era, it was ChatGPT on the Internet. Around Sonnet 3.5 it moved to terminals and encoding agents in the terminal. Around February, GPT-5.3 began making its way into desktop applications. And now we are entering the era of permanent agents. So, this is OpenAI's first step towards permanent agents. As I said, a promising form factor. I think they have a lot of reasons to win because it's integrated directly into ChatGPT, but honestly, from my testing, it's not quite ready yet. It's just that all these little things don't quite work to make this experience great. I'm sure it will improve. So my recommendation is to wait a week or two and then try. I will be at DevDay all day today. I'll be interviewing Sam Altman on this channel, so join me tomorrow for my interview with him. We'll also be releasing a DevDay vlog, so stay tuned for updates. And see you in space. I don't know.