Hands-on early-access review of OpenAI's new Codex app, a GUI for managing coding agents. The video covers cloud-to-local sync, the sidebar for managing parallel agents, the skills library and automations, comparisons to Claude Code and Cursor, and live Reach Test ratings. # Channel Identity Observations: Every ## Explicit Self-Description ### Channel Identity & Positioning - **Name and tagline**: The host describes the channel/company as "Every," with the recurring slogan: *"Every is the only subscription you need to stay at the edge of AI."* - **Host identity**: *"My name is Dan Shipper. I'm the co-founder and CEO of Every."* Joined by Naveen, introduced as *"GM of Monologue... a smart voice to text Mac and iOS app."* - **Claimed scope** — three pillars: *"we do ideas, apps, and training"* - **Ideas**: *"a daily newsletter where we talk through everything new that comes out in AI and give you our hands-on vibe checks"* - **Apps**: a bundled suite — Monologue (dictation), Sparkle (AI file organizer), Kora (AI email assistant), Spiral (AI writing assistant) - **Training**: live streams and subscriber events (e.g., upcoming "Open Claw Camp") ### Signature Content Format - The channel brands its reviews as **"vibe checks"**: *"we always get access to new models before they come out... and we do really good hands-on in-depth reviews of them that we call vibe checks."* - Also references a **"reach test"**: *"our way of telling you did we reach for this model or this product day-to-day"* — with red/yellow/green ratings *"from our engineers who are using this stuff every day."* ### Business Model Claims - **Freemium subscription structure**: *"there's a lot more for paid subscribers"*; free vibe check page plus *"a really really detailed deep dive"* behind the paywall at every.to. - **Early-access claim as differentiator**: *"We get access to this stuff before it comes out"* — from *"OpenAI, from Anthropic, from Google."* ## Observable Presentation ### Intended Audience - **Primarily engineers and technical professionals**, especially those evaluating AI coding tools. The discussion assumes familiarity with Cursor, Conductor, Claude Code, Ghostty terminals, worktrees, and code diffs. - **Explicit audience segmentation** in the review itself: *"if you're currently using codex in the terminal, cursor or any traditional IDE, like this is an immediate switch"*; *"really senior engineers who are already using codex a lot."* - **Secondary audience of AI-interested generalists**, addressed via subscription pitches: subscribe to *"just keep up with AI"* (Naveen's framing). - Live chat audience of roughly **2,000 concurrent viewers** is acknowledged; questions from chat are answered directly. ### Demonstrated Content Scope - **Early hands-on product reviews**: the entire video is a day-zero walkthrough of a pre-release OpenAI Codex app, tested internally *"for the last week or two."* - **Comparative analysis across AI tools**: repeated comparisons of Codex vs. Claude Code/Opus vs. Conductor vs. Cursor, including host's own usage split (*"before... 80% cloud code and 20% codecs and now it's about 50/50"*). - **Internal team expertise as evidence**: multiple Every team members cited as testers (Dan "green," engineering lead Andre "green," Kieran Clen "yellow," Naveen "green"). - **Product cross-promotion**: substantial segments promoting Every's own apps, especially Monologue, including a landing-page showcase and Naveen's usage-analytics demo. ### Presentation Style - **Conversational, casual live-stream format** with humor and running bits — notably a recurring *"stay hydrated"* water-drinking gag framed as a mock sponsor: *"this show is brought to you by water... make sure your agents are hydrated too."* - **Personality-driven**, with informal color commentary (e.g., Naveen comparing Codex to *"that grumpy senior engineer"* you dread talking to but who gets work done). - **Interactive**: host reads and answers chat questions (about Conductor, QA use cases, animation credits), credits staff by name (creative director Lucas Cresbow), and reads viewer comments aloud. ## Claims vs. Demonstrated Patterns - **Claim**: early access to unreleased products — **demonstrated**: the video reviews an app "we've been testing it internally... for the last week or two" before its launch. - **Claim**: in-depth, engineer-validated reviews — **demonstrated**: specific, technical critique of the app (cloud/local handoff, queuing UX, YOLO mode, worktree behavior, automation quirks) grounded in real usage. - **Claim**: monetized subscription ecosystem — **demonstrated**: multiple explicit calls to subscribe, paywalled deep-dive content, and bundled app promotion woven throughout the stream. # Delivery Observation: "We Got Early Access to OpenAI's New Codex App" ## Format & Structure - **Two-host live stream**: Dan Shipper (host, Every CEO) leads with guest Naveen. The format is conversational demo + review. - **Recurring segments**: Dan repeatedly re-introduces the show for late joiners — "if you're just joining us," "if you just got here" — appearing at least five times, each time restating what the product is and linking the vibe check page. - **Formulaic reviews**: Dan names the structure explicitly: "talk you through our testing like what the app is, what we tested, what we think of it, who should use it, who should not use it." - **Sponsor-style break**: Mid-stream he pauses the demo — "I'm going to pause here really quickly and just remind you all AI progress is happening super quickly. Um and you should make sure that you always hydrate" — turning water into a running ad bit. ## Vocabulary & Recurring Phrases - **"Vibe check"** is the signature term, used constantly: "day zero vibe check," "we call vibe checks." - **Repeated framing phrases**: - "really really" as intensifier: "really really smart talented engineer," "really really important," "really really detailed deep dive." - "I think that's really true," "that's really cool," "really slick," "really nice" — positive evaluations cluster on "really." - "one of the really cool things" (used twice about Every's early access and Naveen's usage tracking). - **Jargon used freely**: YOLO mode, unhobble, agent orchestration, GUIs, worktree, diffs, vibe coding, "speedrun" ("we speed ran the terminal era"). - **Slogan repetition**: "Every is the only subscription you need to stay at the edge of AI" appears nearly verbatim multiple times, always paired with "ideas, apps, and training." ## Teaching / Demo Habits - **Live show-and-tell with commentary**: Walks through the UI feature by feature — "So basically on the left you have a list of threads" — defining terms as he goes: threads "are just like conversations," workspace "is sort of the equivalent to a folder." - **Comparison-based explanation**: Anchors new things to known ones: "it looks I think the thing that is closest to right now is actually conductor," "they're like kind of like crown jobs" (cron jobs). - **Direct questions to the guest** to pivot: "Naveen, do you want to introduce yourself?", "What do you think about that Naveen?", "Is there a way that you use this that I don't understand Naveen?" - **Positives paired with caveats**: "I still think cloud code is amazing… Whereas I think Codeex is just like a little bit not like that." ## Communication Patterns - **Frequent disfluency**: Heavy filler — "um," "uh," repeated words ("what what what the app is," "the the the") — consistent with unscripted live delivery. - **Audience engagement**: Addresses chat directly, answers named commenters ("Orion asks," "Mr. Golden Rose asks," "Bronson it is"), asks for clarifications: "I honestly don't know what that means but I would love for you to clarify in the chat." - **Repeatedly praises colleagues by name** — Naveen ("this is why you're a legend"), Andre, Kieran, Lucas — a consistent habit of crediting team members. - **Running humor bits**: Hydration/water ("keep your agent liquid cooled"), teasing about diffs — "diffs [laughter] not beautiful… you got to fight about this Naveen." - **Self-aware asides**: "I'm gonna pause here because I've been blabbing," "I'm biased, but I think it's great." ## Stability vs. Isolated Moments - **Stable patterns**: Filler words, the "vibe check" vocabulary, "really really" intensifiers, re-onboarding late viewers, product CTAs, and question-handoffs to Naveen all recur throughout — well-supported by repetition. - **Isolated**: The hydration bit recurs (three-plus mentions), but individual jokes (MIT of India, Granola rebrand dig, daffodil animation praise) appear once each. No strong evidence these would generalize beyond this stream. ## Uncertainty Notes - Tone is clearly casual-enthusiastic, but "personality" inferences (e.g., whether Dan is naturally promotional) cannot be established from delivery alone — the promotional cadence is observable; motivation is not. - No timestamps or metadata were available beyond the transcript; audience size ("about 2,000 people") is stated by the speaker, not independently verifiable here. # Vibe Check: OpenAI's Codex App — Position Record ## Speakers - **Dan Shipper** — Co-founder and CEO of Every - **Naveen** — GM of Monologue (smart dictation app), identified via transcript self-introduction ## What the Product Is (as described by speakers) - OpenAI's Codex app, framed by OpenAI as a **"command center for agents"**; Dan agreed with that characterization. - **Dan Shipper**: It is neither a terminal app nor an IDE; the closest existing product is **Conductor**. Key features he cited: - Left sidebar of **threads organized by workspaces**; workspaces can span multiple folders/repos (supports monorepo-style organization). - **Skills library** (first-party, well supported); easy to create skills by chatting rather than a wizard — Dan called this "really slick" and "agent native." - **Automations** ("kind of like cron jobs") — scheduled prompts; Dan noted a skill that compounds learnings into agents.md / claudemd files. - **Cloud/local handoff** with worktrees — Dan claimed it "works better than the ones I've used." - **Easy "YOLO mode"** — full computer access toggle. Dan said his prior gripe was Codex was "very hard to unhobble"; he personally couldn't find the right flags despite this being "literally my job." - **Code diff sidebar** — Dan stated he doesn't like reading diffs. ## Positions and Reasoning ### The Terminal-to-GUI Shift - **Naveen**: His usage data (from Monologue tracking) shows Codex usage rising and **Ghostty terminal CLI usage falling**; the Codex app is now his go-to. He said the UI provides context the terminal can't: "you can clearly see the UX of how you are able to discover new features... you can't really get that from CLI." - **Dan Shipper** interpretation: Naveen is "a leading indicator" for a market segment; "we might have speedrun the terminal" and GUIs are coming back — but "the GUIs we were using were just built for a different era of coding." He later qualified: terminals will still be around, but "we probably won't be doing most of our work in them pretty soon." ### Disagreement: Diffs - **Naveen**: Loves the inline diff feature; for precise changes and "really tough bugs," he checks diffs "to make sure it's making the edits as I would do." He also said for big features (e.g., building the Monologue iOS app) he looked only at high-level architecture, not code. - **Dan Shipper**: Disagreed in part — "you probably shouldn't be reading actual lines of code, usually you should be reading summaries." He conceded plan review differs from diffs and "sometimes you need to go down into the lower level." ### Model/tool Split: Claude vs. Codex - **Dan Shipper**: His usage shifted from ~80% Claude Code / 20% Codex to ~**50/50**. Claude Code (with Opus) is faster, "more empathetic," has "better taste," better for non-coding tasks and UX planning; Codex is slower (e.g., starting a server) but better for numbers, math, performance-critical work, and bug-free production code. - **Naveen**: Agreed — he didn't switch to Opus as his main model; uses Codex/GPT-5.2-Codex for engineering. He avoided Anthropic's Opus for creative/new work ("I don't want to give it to Codex because it's not that creative"). He cited a tweet from the OpenClaw creator (referenced, not verified) that Opus isn't allowed in his codebase while Codex works on actual code. - **Shared metaphor**: Codex is "that grumpy senior engineer" — not fun, but honest and gets work done. Naveen: "Codex reviews... completely honest... not going to glaze you." ### Cloud Collaboration Limitations (Dan's gripes) - **Dan Shipper**: Cloud tasks spin up/down environments; follow-ups spin up new environments — "not easy to collaborate." He claimed the Claude cloud app handles this better, interleaving thinking and chat. Also: queued follow-up messages either wait or stop-and-restart the agent, whereas Claude evaluates whether mid-run messages matter and continues. - **Naveen**: Confirmed he had the same issue ("I had the same issue"); his workaround is dictating very complete task prompts upfront to avoid back-and-forth. ### Conductor Comparison - **Naveen**: Was "all in" on Conductor before the Codex app; moved away mainly because Conductor **always creates worktrees**, which feels heavy for quick tasks and conflicts with Xcode workflows. He said the Codex app "got heavy inspiration from Conductor for sure" (guess, stated as such: "my guess is"). - **Dan Shipper** reaction on seeing Conductor: "It really is very similar." ### Atlas Skill Limitation (live demo) - **Dan Shipper** demoed the Atlas skill; it could manage browser tabs but couldn't read page content. He characterized this as typical of Codex being "hobbled" by security gating — "harshes the vibe" — and said Anthropic is "generally a little bit better at opening things up." He explicitly flagged uncertainty: "this might not be actually the Codex team's thing. It might be that the Atlas browser itself doesn't allow you to do this." - He also noted an automation whose function/success he couldn't verify: "how would I know if this worked?" ### Who Should / Shouldn't Use It - **Dan Shipper** recommendations: - **Immediate switch**: current terminal Codex users, Cursor/traditional IDE users, and senior engineers who like diffs and want agent orchestration. - **Worth checking**: Claude Code users dealing with "nasty production bugs or gigantic code bases." - **Not recommended yet**: vibe coders / weekend app builders — "still a little too slow," "still has no personality." He did endorse using Codex as a sub-agent inside Claude Code for plan review/code review. - For QA specifically, he suggested Claude with browser may currently be the better option, though Codex is good for writing tests/fixing bugs. - **Market positioning**: Dan observed OpenAI's marketing is shifting Codex toward everyone, but "the reality of the app as it is today is that it is still actually mostly for people doing... really high performance coding." **Naveen** agreed Codex targets senior engineers, noting a new ChatGPT "apps" experimental feature connecting to Notion etc. signals a mainstream-work direction. ### Team Reach Test (Every internal) - **Dan Shipper**: Mostly green — himself (50/50 with Claude Code) and Andre (engineering lead) are Codex fans; **Kieran** (GM of Kora) rated **yellow** — to him the app "still feels like a step backwards." Dan: "I'm actually very curious what he'll say in like three or four months." ## Unresolved Tensions / Qualifications - Dan's GUI-revival thesis coexists with his acknowledgment that terminals "have been around for 40 years. They'll still be around for a long time." - Naveen's "100%" agreement with Dan's summary-over-diffs position is qualified by his own continued diff use for hard bugs. - Dan praised the cloud/local handoff as working well while separately criticizing cloud task collaboration — both positions held simultaneously. - The Atlas limitation's cause (Codex gating vs. Atlas browser design) was left explicitly unresolved by Dan. # Video Attribution Record ## Video - **Title:** We Got Early Access to OpenAI's New Codex App: A Command Center for Agents - **Format:** Live stream / "vibe check" review hosted by Every (every.to) ## Speakers & Roles ### Dan Shipper — **Host / Creator** - Self-identified: co-founder and CEO of Every. - Primary narrator throughout; conducts app demo, presents the vibe check, reads and answers audience questions, and delivers promotional segments. ### Naveen — **Guest** - Self-identified: GM of Monologue, a "smart voice to text Mac and iOS app." - Guest commentator; provides hands-on demos (usage graphs, Monologue codebase workspaces, Conductor comparison) and answers host questions. ### Referenced non-speaking third parties - **Kieran Clen** — GM of Kora at Every; cited as tester with a "yellow" rating (did not adopt the app heavily). - **Andre** — Every's engineering lead; cited as "green" tester. - **Lucas Cresbow** — Every creative director; credited with the vibe-check page animation (named in answer to an audience question). - **Nat Elias** — announced guest for an upcoming paid-subscriber "Open Claw Camp" event; not present in this stream. - **"Petito" (per transcript)** — cited by Naveen via a tweet about not letting Opus in his codebase; third-party claim, unverified. ## Promotional Segments (Flagged) - **Water/hydration bits:** Recurring promotional-comedy segments ("brought to you by water," four glasses of water, "liquid cooled" agents) framing Every's "stay at the edge" branding. - **Every subscription promotion (multiple):** Host repeatedly promotes every.to as "the only subscription you need to stay at the edge of AI," including paid-subscriber deep dives, red/yellow/green ratings, and reach tests. - **Every apps promotion:** Monologue (monologue.io), Sparkle, Kora, Spiral bundled subscription. - **Open Claw Camp promo:** Paid-subscriber event Friday with Nat Elias; explicitly described as a "sneak peek world premiere" announcement. - **YouTube/subscribe CTAs** repeated for new joiners. ## Referenced Sources - **OpenAI blog post:** "Introducing the Codex app" at openai.com (URL garbled in transcript as "openai.comindexintroducing thecodex app"); shared in chat. - **Every vibe check page:** every.to (URLs garbled as "every.vivecodex" / "every.TOVEC-VIE-check"); free and paid tiers described. - **Third-party products discussed/demonstrated:** Codex app, Conductor, Claude Code, Opus 4, GPT 5.2 Codex, Ghostty, Cursor, Atlas skill, Linear skill, Open Claw / Clawbot, Figma skill. - **Third-party claims:** Naveen cites an unnamed tweet about Opus vs. Codex usage; host references Anthropic-internal framing ("unhobble") without a cited source. ## Audience / Comments - **Live chat active:** Host states ~2,000 viewers; reads questions by name (Bronson, Orion, "Mr. Golden Rose," Joe, Colleen, Austin, unnamed follow-ups). Comment availability: **available during stream**; some questions answered on air, one deferred ("I honestly don't know what that means"). ## Provenance Notes & Uncertainties - Speaker turns are only partially marked by `>>` in the transcript; attribution of some short exchanges inferred from context and dialogue flow. - Name spellings uncertain from audio: "Naveen/Navin," "Petito," "Kieran Clen," "Lucas Cresbow." - Several URLs and product terms are garbled in the transcript; recorded as stated, not corrected. - No timestamps available in the source transcript. All right, everybody. Welcome to a vibe check of a new product from OpenAI. It is a codeex app. Uh we've been testing it internally here at every for the last week or two and we have your day zero vibe check. This is the OpenAI blog post. You should check it out at openai.comindexintroducing thecodex app. We will throw the blog post in here. Uh Um, I just threw it in the chat for all of you people on YouTube. Um, we also have a vibe check on every. So, one of the really cool things at every is we always get access to new models before they come out and new products before they come out. And we do really good hands-on in-depth reviews of them that we call vibe checks. So, I'm posting a vibe the vibe check link over here. You can read it. I'm going to go through it on the live stream. Basically what we're going to do today is talk you through our um uh our testing like what what what the app is, what we tested, what we think of it, who should use it, who should not use it. Um and then uh uh and and all this is also on our website at every.vivecodex. You can check that out. Um my name is Dan Shipper. I'm the co-founder and CEO of Every. And I'm joined by Naveen. Naveen, do you want to introduce yourself? Hey everyone. Uh, I'm Navin, GM of Monologue. It's a smart uh, voice to text Mac and iOS app. So, I'm really excited about the new app. \>\> Excited to to have you here. So, um, let's let's just talk about what this is. I'm going to actually go to the OpenAI blog post really quick. Um, the way that they talk about this on the OpenAI website is that it is a command center for agents. And I think that's really true. Um, when we tried the app, and I'll open it up here. Uh you should be able to see it here. Um this is what it looks like. So it's um it's out of one of the thing one of the things you'll notice immediately is it's out of the terminal. It is not a terminal app. Um but it's also not an IDE. So it like does not look like cursor. Um it looks I think the thing that is closest to right now is actually conductor. Um so basically on the left you have um a list of threads. Um, and threads are just like conversations that you're having with with your agent with codecs. Um, the threads are organized by workspace. So, um, inside of workspace, this is sort of the equivalent to a folder inside of your inside of your, um, desktop, but you can work, I believe you can work actually on multiple folders at once inside of a workspace. So, if you have multiple repos, you can organize them into a monor repo in here. So, that makes that really easy. Um, so I have a couple different projects in here. One is proof, which is this agent native markdown editor I'm working on. This is our our growth OS is like basically how we track growth inside of every um so that's the left sidebar. It also has access to skills. So skills are like a um firstparty like really well supported thing inside of inside of this app. It comes with a skills library with some pretty good um pretty good skills by default. And you it's also really easy to make a skill. So you can just like click to make a skill. And what what I love about this is this whole this app is very agent native. So instead of it having like a separate wizard that you can use to um to to go make a skill, you actually just chat with it and say, "Hey, I want to make a skill. It can do can you make it do this?" And that just works, which is I think really slick and really nice. So we've got skills. We've also got automations. And these are essentially like they're like kind of like crown jobs. You can set them to um to have codecs run a prompt on a schedule. So you can see them sort of building a little bit in the kind of almost clawbot or open claw territory here where um you know I have a skill that just compounds the learnings from my um from the development I've been doing into my agents MD file and my cloud MD file and that just like runs automatically. Um and yeah, so that's the that's the left sidebar. Um then what we have if we go in here, this is just an agent chat. Some of some of the cool like bells and whistles that you've got here are um one is uh you can change whether it's uh going to cloud or local. So there's really good uh local workree and cloud handoff which is amazing. Um it actually works pretty well. Um and this is one of the things that uh a lot of the other um uh model uh agent harness providers are working on. But I this is this one actually works better than the ones I've I've used. Um, Codex also has this like really easy way to just go into YOLO mode, which that was like one of my big gripes with Codex is it felt like it was um made for um uh it was it was very hard to unhobble it as some of the folks at Anthropic call it. And this just lets you be like, hey, you can have access to the whole computer and that's really great. It does a lot of really good stuff. Um the last part of this is it has a um it has a sidebar that shows code diffs. So if you're someone that likes looking at code diffs, I certainly am not one of those people, but if you are one of those people, um uh it it has that. So that's like the really quick uh that's a really quick tour. I'm going to pause here really quickly and just remind you all AI progress is happening super quickly. Um and you should make sure that you always hydrate. So I've got four glasses of water here, um of varying sizes, and this is really \[clears throat\] what you need to stay at the edge. So, if you're interested in staying at the edge, make sure you're make sure you stay hydrated. And also make sure that you subscribe to Every. Every is the only subscription you need to stay at the edge of AI. Um, we get access to this stuff before it comes out. And we have vibe checks like this one on our website. You can check it out at every.TO. Um, you can read read through it. You can chat with it with Claude. Um, we have um, you know, red, yellow, green type uh, ratings from internally from our engineers who are using this stuff every day. And there's a lot more for paid subscribers. I'm gonna pause here because I've been blabbing and I want Naveen, I want you to give me your take. So um you use you use Codex all the time prior to the app coming out and uh like what what do you think of it now? \>\> Yeah. Uh so I think I'm one of the few users who didn't like you know engineers who didn't go to Opus 4 uh as a main model. I still use Codex GP 5.2 too. And what happened is just before the Codex app release, I started going all in on Conductor. Conductor was like really great app for me where you know I need that UI to manage multiple projects and then but one thing I really like you know not like about conductor is it's always creates this workspace and as a iOS/ you know Mac developer when you're dealing with workspaces uh it's like really hard uh with the next codes uh projects. So which I not like it. So but when Codex realized data and then they have this like you know individual threads locally but you can also create work trace you can also do cloud uh I was like super excited about that and yeah I'm like maybe I can show you guys like what I did. Uh \>\> yeah please \>\> absolutely \>\> ju just to set a quick uh you know context before Codeex uh like you know sorry uh the app happened. This is my let me show you my monologue usage. It's quite low recently because we had a think quick last week. So my total number of words decreased. But here you can see codex has been my like you know go to app and the usage has been increasing a little bit and ghosty which is my terminal uh CLI where I use codeex and then it's decreasing. So this is like the graph that I pulled quickly. Here you can see initial like blue line I'm using go a ton but slowly I started like you know interacting more with codex a lot and and now I think codex is like my go-to app instead of ghosty. Uh so that's the quick context. \>\> Wait wait wait let me just let's move let's go back there. So go back to the graph. \>\> Yes. \>\> This is one of the really cool things about Naveen is because he's building monologue which is a smart dictation app which is bundled in with every. So if you are if you want uh if you want to get monologue you want to get every just go to every.0. You can get monologue or just go to monologue. Is it monologue.to? \>\> Yeah, monologue.io. We don't have the \>\> monologue. Sign up for that directly. It's really awesome. It's just like it's the best smart dictation app. I'm biased, but I think it's great. Anyway, one of the really cool things about it is he can track what he's using and I think this graph is actually very very important because I think Naveen is a leading indicator for where the market is going in a lot of ways um for a particular segment of the market. So Naveen is one of the few people inside of every that is not has not been like super cloud code pill. Like you use cloud code but like you've for for monologue in particular because it's um I mean you're you're just like a really really smart talented engineer and a lot of the stuff you're doing with um monologue is highly complex production code that needs to run really fast and be really performant. Codex is really good for that. So you were using codecs before, but what's really interesting is you were using codecs in the terminal in Ghosty. And what you can see is when we got access to this um uh when when we got access to this app, your usage flipped from the blue, which is ghosty in the terminal to the app. So it like it succeeded in basically um uh getting you getting you to switch out of the terminal. That's I think very very important because we're living in this world right now where because of cloud code for the last six months I think we've all moved to the terminal and using UIs has felt pretty backwards actually. Um but this what this says to me is we might have speedrun the uh the terminal and we're now moving back into into UIs and and gueies are maybe actually like they they still are good. just the gooies we were using were uh just built for a different era of coding and now we finally have some people starting to build these kinds of things that are really for this like conductor and now the codeex app. What do you think about that Naveen? \>\> 100%. Like I'm still a fan of a CLI. It's easy but again it's just giant wall of text right in terminal. So if you're like I'm not really terminal user. I I love working with UIs and I'm like a product person but engineer as well. So when the like you know codex app came and it's like really beautifully designed I'm like went all in and I really love the use uh user experience of using an app. So for example, let's say if you're using a CLI, you have these tabs, right? Like for example, I have multiple code bases here. It's all in sidebar. So I can always see what's happening within terminal. You can't really get that kind of experience. So I I'm bullish on UI UI \>\> because kind of gives that extra context that you can't really get from terminals. \>\> Yeah. Actually, can you explain that in more detail? Like what's the thing that you're getting here that you it's hard to get in terminal? \>\> Yep. Yep. Okay, let let me give you a base context where I have monologue uh like four three different reports. one is my go-er which is like the high like you know high lat low latency ton of traffic gets through this and then I have the main back end which is in python and then I have like my iOS macos like monlock clients right so all these three are like three separate code bases but what's really cool thing is uh I can create a workspace by adding all these three into one single \>\> workspace so that's what I did mono code this has like access to all my three different uh code bases which is really cool. Now let's say if I'm working on a feature that touches all three different code bases, I just go like create here and start like you know monologuing into this and then it task gets done. But now let's say today I'm like working on individual task within my main back end where I'm trying to update a API. So here now I can just come here start creating tasks and then you get this beautiful like you know diff on the side which is really cool \>\> diffs \[laughter\] not beautiful. \>\> I love \>\> you got to fight about this Naveen. \>\> Tell me why it's beautiful. \>\> Yeah. So for me uh when I'm like you know uh doing any big code uh like new features I don't really look at uh depths. But when you're doing this like precise changes for example in this particular case uh you you see this uh analytics right it's just lost 30 days but I want to see also my lifetime usage of monologue so I'm thinking oh what can I do I just monoged it and then I got this plan in line now I'm like reading this and I'm like it has few questions and I can just like say immediately leave few comments like saying oh this is not good or thing and then here you can see the comment get attached immediately and I can just like let agent again look at all my individual comments. So this what this enables you is you can now do this precise edits of each individual plan or individual code. So you're kind of like not letting agent run completely but you're also giving you feedback giving feedback and then iterating. Uh so I'm like little bit much more you know what do you say u much more with the agent than my previous CLI usage where I'm always like going back to different app to read the markdown and then come back and then give feedback like to and fro now \[clears throat\] all I'm just like all in in codeex. \>\> Okay. Yeah. I mean I agree with that. I think um this doesn't count as a diff really because this is this is a plan. It's a bit different. Yeah, \>\> I I I I'm I'm all for like reading a plan and reading updates to a plan. I just I think what I'm where I'm going with this is when you're at a point where you're orchestrating agents to like do a lot of code for you. \>\> You probably shouldn't you should know what they're doing, but you probably shouldn't be reading actual lines of code, usually you should be reading summaries of what they did that you can that like tell you where you need to look. That's that's my feeling about that. Sometimes you need to go down into the lower level, but I think over time we're going to do less and less of that and more and more of diffs of here's what I accomplished and the the important details that you need to know so that we can check to make sure it's right. What do you think of that? \>\> Yeah, I'm 100%. I think recently as well when I'm building out the whole iOS app of monologue, I didn't look at the code. I just looked at the highle architecture and obviously I asked Codex to give me the you know uh how the architecture is set up everything. So I I also stopped looking at code. But when I'm working with like like really tough bugs, then I really look at the diff and see make sure it's making the edits as I would do so that I still have a good understanding of the bug fixes that you know we are making. So \>\> that that makes sense and I think everyone should know that um Naveen went to the MIT of India and so he's super smart. He's much a much more talented engineer than me. Um, and so I think one of the things that we came to, um, when we when we do our vibe checks, what we like to do is try to figure out, um, who is this for? Uh, who's going to really like this and who's who who might not like this. And, um, for this particular release, um, let me just see if I can open this up in our, uh, uh, here we go. I'm going to open it up. I'm going to share my screen. So um on every so we have basically this um free and open vibe check page here which has our highle take and then for paid subscribers on every we have like a really really detailed deep dive into everything about the app all the testing that we did a walkthrough of it all that kind of stuff um a reach test which is also public and then a bunch of other stuff that I'll go through in a little while but um what's really interesting is um when you think about who it's for and how it stacks up, I think the thing that we came to is um if you're currently using codeex in the terminal, cursor or any traditional IDE, like this is an immediate switch. Um it's uh uh it's a it's a guey for programming that is not an IDE um but that is built for agent orchestration and that's actually really powerful and it's like an immediate um it it immediately became part of my workflow. I'll say that and I've been I've been using I would say before I started with this I was using probably 80% cloud code and 20% codecs and now it's about 50/50. I still think cloud code is amazing. Um it's especially with Opus, it just it's a lot faster. Uh I I find it's like more empathetic. It has better taste. Um it's better for tasks that are beyond coding. And it's better for like okay, I need to make a plan for a really delightful user experience like that kind of thing. Whereas I think Codeex is just like a little bit not like that. Or for example, if I want to start a server, I use I typically use cloud because like will take 30 minutes to like start a server currently. Um uh but uh if if you're like a if you're a senior engineer who really likes to look at diffs like they made this app in a way that should be an immediate yes like I want to try this and I want to step into like real agent orchestration with codecs. \>\> Yeah 100%. Uh yeah I agree with what you said even uh my usage of opus is always uh oh we doing some something new creative thing like I I don't want to give it to codeex because it's not that creative. \>\> Yeah totally. Um okay let's let's like move on. I want to show you guys a couple other things. So um I want to do the cloud to local sync. You can see my screen right? Yeah. So, um, one thing that's really cool about this is I can click cloud here and I can kick off a task. Like, um, what's a good task? Uh, right now there's, um, a problem when proof, this is a bug, a real bug I dealt with recently, when proof loads um, MD files that have YAML in them, like skill files. Um, I should be monologuing right now. Um, open it up because I closed it for a video I was recording. \[snorts\] Um, let's see. Come on. Uh I want to make sure that uh that that it it doesn't corrupt that and that the YAML actually shows up inside of proof and is fully editable and doesn't inter it doesn't interfere with the uh with all the proof provenance stuff. Yes, Bronson it is. Um and so what I can do is maybe I'm like going out for a walk or like I'm going out to lunch or whatever. I can just click cloud. One of the things it does it does the parallel things which is really cool. Codex has had that for a while in the web app and cursor also has this and I just click go and \[snorts\] uh it just fires it off inside of codeex the uh and and I can go access that. Let's see if I can open and access it. The gripe that I have with this is and I think this actually connects to like a oops a larger my P button is not quite working. A larger gripe I have with um the way that the codeex model is oh oops the way that the codec model is structured is I I I actually asked I kicked this off uh a couple days ago uh because it's it's a real bug and what happens is you get a response but it's not actually easy to like chat with it. it like spins up an environment and then spins down the environment and then if you want to follow up it's like spinning up a whole new environment and it's not easy to collaborate and I think uh the cloud app does a much much better job of this where um it is both an agent that can go off for a long period of time and it will just respond immediately with a question that you have and and I think that kind of a flexibility and collaboration is really important. Is there is there a way that you use this that I don't understand Naveen that makes this work better for you? \>\> Uh actually I had the same issue. I think when Codex cloud I think these guys uh implemented Codex cloud when it's in April I think last year 2024. So the way they implemented it like what you mentioned uh but the cloud code one the cloud the cloud version that they have it's like just something similar to what how you use it in CLA right it's just like easy chart it and then I think they need to update codex needs to get updated for me what I do is I just monologue a ton and make sure like it's a really proper task that I can just hand it off instead of this two and fro to and fro collaboration that usually happens with CLA tools. \>\> That makes sense. Yeah. I Yeah, I think that's the that is probably the right way around it. The other thing that's that's interesting is um I want to show you another thing that I think is is a limitation here that I really hope they fix soon is if I say, "Okay, hey, I want you to fix this bug where if we load a file that has YAML in it, um it corrupts the uh uh it corrupts the proof propance at the bottom. Can you look at that? So, if I say this and I move it to kicking it off in on on local and we're in, you know, full full access mode, which is really cool. If you should definitely press this button if you're if you're using this app for the first time. It's it's really nice that they added this. It makes it feel like, okay, I can get into YOLO mode really fast, which before like I know Codeex had this, but and this is literally my job to know this, but I had no idea how to do it. I like tried a few different flags and then just gave up and just accepted permissions all the time. Anyway, if I do this, what's kind of interesting uh or what what I don't like about this is it's thinking it's starting to list files and I'm like, "Oh, um I realized that hey, uh I realized that I need it um the YAML to be editable um by the user in the document. So, it should show up there too. Not the proof YAML, but anything else." So, I want I basically realized something as I as I put as I kicked it off. Um, if I press send, it cues it. Um, which is not necessarily what I want to be honest. And I think you can turn off the queue. Yeah, you can turn off queuing, but if I if I then push it, it's going to stop it. It's stopping it in the middle of what it's doing and then restarting it. And what Claude does is it just like it takes a look at it. It's like, does this change anything for me? And if it doesn't change anything for it, it'll just like ignore it for a while and like finish what it was doing and then it will um uh start to do your task. And I just think that that UX is much better and there's this sort of interled thinking and chatting thing that it can do that codeex does not do yet that I think is really really important for this kind of thing. \>\> Yeah. And also here you also given the case right cla versus GI like you can clearly see the UX uh of how you are able to discover new features like oh queuing versus uh immediately stopping the chart that you can't really get it from CLI you \>\> yes yeah CLI is a little bit harder because yeah like there's a flag for this that I would have to find or like you know discover that yeah definitely definitely the UI is different here Um, if you're just joining, uh, what you should know is that OpenAI just dropped a new codeex app. Uh, we have the full analysis, the full vibe check of the codeex app on every.TO. We've been testing it for about a week now. We put it through its paces. Um, this is the kind of thing that we do all the time at every. We get access early before they come before things come out and we do a vibe check. Some of the things that are interesting about this new product is it's a UI for coding. Um, but it's not an IDE. It is actually an agent orchestration um application that's very similar to Conductor that lets you kick off and manage multiple agents. It's got a sidebar that lets you uh do parallel agents all at the same time. It's got cloud to local sync. It's got great automations and skills which is quite cool. Um it has code diffs. Um we've got the live stream here. We also do this thing at every call the reach test. Uh where we uh try to figure out are is this something that we reach for? Um, I think that's that's like the the key metric for any AI uh new AI model or new AI product is like does this actually make it into our workflow? And the answer is yes. At least for most of us. For me, I would say it's like about 50/50 with cloud code now, whereas before it was 8020 with cloud code. Um 80% cloud code, 20% uh codecs. Um I think what what it did for me is be like, oh, actually there's a future for UI here. It's really nice to have guey. Um, I had thought the guies were over, but it turns out I think terminals are maybe they're not over. You know, terminals have been around for 40 years. They'll still be around for a long time. But, um, I think we'll probably won't be doing most of our work in them pretty soon. So, I was agreeing on this, which means I use it almost every day. Andre, who is our engineering lead at every is also green. He's also a codec stand. Naveen, do you want to talk about your take? \>\> Yeah. Uh, I my my take I'm super great. I'm using Codex a ton and for me what makes uh like why I'm featuring to codeex app over CLI or opus is the sidebar so that I can like you know keep track of all the things that I'm working right now. That's like super important because I've been recently struggling with you know focus or the things that I'm working on currently. uh because let's say if I kick off a task in my back end now I'm moving to Mac app or iOS app it's getting difficult for me to manage all the things that I'm working on I think ton of people within our team also little bit struggles of okay what are all the things I'm working on currently and you know codeex have made it easy for and then obviously uh you know uh that's like one key thing and then yeah uh the diff inline diff that's like my favorite feature because uh the Codex app is kind of like reimagining the IDs, right? So, they're able to like do it really well, I think. \>\> Totally. Um the one person on our team who tested this, and usually have we have a few more testers, but we're pretty busy right now. Um is Kieran Clen, who's the general manager of Kora, the uh he's the guy that came up with Compound Engineering. Uh he's amazing. He's super cla and he's a yellow. Um I know that he's been using the codeex model with Claude, but he's not using the app that much. To him, it still feels like a a step backwards. I'm actually very curious what he'll say in like three or four months. Um but I I do understand why. And I think particularly one thing that you should you should think about is it who should try this. Right now for us, it's like really senior engineers who are already using codeex a lot or already using something like codeex. Um, or maybe you're using Cloud Code and you're uh finding that you're using it for like nasty production bugs or in gigantic code bases. It's it's worth it's worth checking out the Codeex app. If you're vibe coding all the time, I do not recommend this yet. Um, I think it's still a little too slow. it still it still doesn't feel like it's meant for that use case, but you can tell that they're going in that direction, which is really good because um it's been about 6 months or nine months since Cloud Code came out and we've been seeing them figure out in real time this this new way of interacting with agents on your computer where the agent has access to everything that you have access to. It can do anything you can do and you're pushing the agent to like go off and do stuff and then coming back and like reviewing its work. That's something that Anthropic has really figured out. And what they've figured out from there is that once you have that, it's a great way to get other types of work done, not just like really difficult programming tasks. It's like great for um any kind of knowledge work. And that's where, you know, tons of people on our team now who are non-technical are using cloud code all the time. And that's where co-work came from, all that kind of stuff. \>\> And you can tell that OpenAI is moving in that direction, which is really cool from the fact that they have a skills library here. That's like one of the core things. They have an automations uh section here. And so you can basically tell um tell Codeex, hey like uh run this prompt every XYZ number of times uh per day or per week or whatever. And that's that's a really cool way to start automating more and more work on your computer. I think another sign of that is this button now CEX has full access to your computer. um before it was like kind of hidden and Codex felt very sandboxed and very like riskaverse and I think uh in order to get it adopted by like more users they need to they needed to change that and it looks like they have but it's still like way too slow. It still has no personality. Um so it's not something that I recommend if you're like just vibe coding apps on the weekend to like really use. You can use it sometimes like um some people I've seen in the chat uh h have talked about using codeex as a sub aent inside of cloud code and that's a solid option to like review your plans or do code reviews. I really like that use case but I wouldn't necessarily switch over to this. What do you think Naveen? \>\> Yeah, I agree with you. uh like there's I think there's clear division happening right slowly with where opus is going and or anthropic is going and then where uh open air wants to go with codeex codeex is like targeted towards uh you know enterprise code or where you need to fix those kly bugs I think even recently yesterday I saw a tweet from Petito who made like you know open claw he said he doesn't let us in his codebase at all and then it says codeex uh working on the actual code and the same thing like you mentioned but actual open claw that's like you know clawbot uh under the hood just uses office right it's really easy to talk with it's just fun but when you're using codeex it's not fun at all so yeah I think everyone is kind of \[laughter\] \>\> it's just like you know that grumpy senior engineer that you always have in your company \[laughter\] that you don't want to talk with but that's the only way to get work done \>\> nine this is why you're a legend. You're like, it's not fun at all, but it's like I use it 100% of the time. Um, like I don't know if you guys know know this, but uh so Naveen is in New York right now and he's staying at the every office and I just whenever I come in I've I've said this a few times in different settings, so I'm sorry that I'm outing you again publicly, but um \>\> whenever I come in in the morning, it's like 9:00 a.m. and Naveen is like at at the kitchen table like usually on the phone like talking to the design team or or someone trying to do something for monologue. And this man does not get up until like six. Um, and six is like probably early for him. And uh, he's just like grinding away all day with Codeex. And that's why Monologue is this app built basically just by him that is competing with teams of like 50 people or more that have millions and millions and millions of dollars raised. And um, uh, and that's because he's willing to use something that he thinks is not fun. Um, \[laughter\] as much as we love fun here at every, I think that's a really energy thing. \>\> Yeah. \>\> So, yeah, it's the same thing in real life as well, right? You have that like senior engineer who's really good at what he does, but you always like dread going to him because he kind of like pulls you apart. \[laughter\] It's kind of like So, yeah, it's Codex is like that. That's why I love Codex reviews as well because it's like completely honest. It doesn't like, you know, try to mince words. So, \>\> it's not going to glaze you for sure. \>\> Yeah. \>\> Um, yeah. Uh, let's see where else we have here. Go for it. Go for it. \>\> There's a question from one of the comments is like what is conductor and how is it different like you know people wants to like do do you use conductor? Did you use conductor then or \>\> I never really used conductor but I would love for you to talk through that because I know that you were and maybe still are a conductor guy. Like how does this work? How does this compare for you? Yeah, I think I moved once codec got codex got released I moved away from conductor but before I was actually using it. So let me do a quick share. \>\> Great. I personally think Codex is the pioneer uh in this uh like UI where they're able to figure \>\> sorry conductor conductor is the like first guys who figured out workspaces where you can add uh you know create folders and then start using I think I yeah I cleared out all my previous things but let me see if I can get something back. So, it's almost the same UI, right? You can see you have the files in the sidebar, you have the terminal access, and then you can just have multiple like \>\> Oh, wow. It really is very similar. That's funny. \>\> Yeah. Yeah. I \[snorts\] think uh they saw my guess is codeex and cloud code people are seeing what's happening with conductor and then like \>\> took the same ideas. You can it's almost the same thing. You have the changes in the sidebar. What's really cool is actually let me see if I can create a new new task. So whenever you create a new thread, it's kind of creating the workspace. You're not actually working uh work tree. Sorry, it's creating a work tree. So you're not actually working on the main code base, but you're kind of like working on a work tree. And once you like let's see can you create a plan uh so that we can add uh lifetime analytics to the current analytics endpoint. So what's really cool is you can change the opus and you can enter like they have UI for everything that's already available in the CLI. For example, right now I'm able to select cloud code opus 4 point and enter into the plan mode. Right? this is something and then you can select the thinking tokens and everything as well which is really cool and I can just send and once it's done uh you can also review it directly within the app so conductor is kind of uh you know trying to replace cursor you can think of \>\> so if if you're just joined someone just asked hello what is conductor near the appal logo we're we're demoing an app called conductor that is we think similar to the way that they built um the Codex app like it just a lot of the UI and the patterns remind us of CEX and Naveen was a big conductor guy and now I think he's moved a lot of his workflow to the Codex app and so he's just giving us a refresher on what conductor is. \>\> Yeah. And one of the main reasons why I moved away from conductor is just because of the work trees how how heavy it feels. Like for example, if you have a quick task, you you can still work on the main main tree, right? So, but for now conductor the way things work, they always create a new uh work tree. And then I just kicked off a task. I think it will take some time, but you can see the UI everything it's like super interesting and you have everything access and you can also have like they have ton of features. I don't want to get into all the features, but uh Codeex I think got heavy inspiration from Conductor for sure. \>\> Yeah, makes sense. \>\> Um I want to take a break to reintroduce if you're just if you're if you just got here, reintroduce you to who we are and what we are doing. So um we're every we are the only subscription you need to stay at the edge of AI. We do ideas, apps, and training. On the idea side, we do a daily newsletter where we talk through uh everything new that comes out in AI and give you our hands-on vibe checks. That's what this is. Um for example, today obviously the new codeex app launched and we had a vibe check the minute it launched with our deep dive review for paid subscribers of um what it is and what we've found, who it's for in our hands-on testing. We get these kind of things early. We get new models. We get new products from OpenAI, from Anthropic, from Google. And so you'll know on the day that it comes out if it's worth your time and attention and who it's for and if it's any good. Um, so at at every we do ideas, apps, and training. So that's the idea side. On the apps side, we have a suite of apps that we built. The first one is Monologue, which is built by Naveen here using codecs. Actually, Monologue is a smart dictation app. It's awesome. You should check it out. Second one is Sparkle. It's an AI file organizer. It's sort of like a little file agent on your computer that um organizes your desktop or any folder you want according to rules that you choose. We also have Kora, which is an AI email assistant. It's an AI agent that sits in your email. It's pretty awesome. And then we have Spiral, which is an AI writing assistant. And this is all bundled together. So you pay one price and you get access to everything that we make. So that's the uh ideas side in ideas, apps, and training. So uh the last one is training. So, we do we often do live streams like this one um but for paid subscribers. So, for example, we have uh something coming up um uh on Friday called um Open Claw Camp. Let me see if I could find us uh some of the early designs for Open Claw Camp because they're pretty they're pretty cool. Um so, this is sneak peek world premiere. This is what the Open Claw Camp uh looks like. Um, we've got uh Nat Elias coming. We'll be announcing this later today or yeah, probably later today. So, sorry Austin, if you're having a having a panic attack that I revealed this, but it's coming up later. It's coming up on Friday. It's going to be really awesome. We're going to uh with paid subscribers, we will uh spend 90 minutes with you with you with Nat Elias and the every team talking about how we use it internally and help you get set up if it's something that you've heard about and uh are looking to uh learn more power tricks of the tricks of the trade power user tips for uh open claw uh open claw and we do these kind of camps all the time. It's for paid subscribers. If you're into this every hit subscribe or check out monologue monologue.to O, it's got a It's got a Let me show you the Monologue landing page because it's so beautiful. And Naveen has a really cool um iOS app coming out soon for monologue um that you should check out. Um it's the shortest distance between talking and typing. I love monologue. Naveen, anything you want to add? Why should people subscribe to every? \>\> Yeah, like you get like first thing you uh you're able to get access understand immediate like what's happening with codeex, right? You don't have to you you can just subscribe to every to just keep up with AI. Even personally sometimes I feel I'm too behind and I just go to every and then you know during the weekend and just start uh reading see catching up. So yeah I think every and you can get to have access uh to all these great apps uh under one single subscription. So that's a crazy deal for me. \>\> I love it. It is a very good deal. The uh the last thing is this uh show is brought to you by water. Um the the edge of AI is always moving and it's really important to stay hydrated. So I have um not one but four glasses of water here and I'll be drinking these throughout the throughout the day. And that's really how I stay at the edge is I make sure to stay hydrated. So you should stay hydrated and make sure your agents are hydrated too. Um cheers to that. \>\> Oh, it's so refreshing. New York New York tap water. No, Naveen, you have do you have a water bottle? Yes, I have like this one. \>\> Oh, see this is why Naveen is so good at what he does. He just has a gigantic water bottle that he brings around. If I was as hydrated as Naveen, I feel like I would be more \[laughter\] productive. \>\> Um, if you are just joining us, welcome. We are talking about the codeex app. Um, it is the new app from OpenAI. It is a app uh that allows you to or it's an a it's a command center for agent. It um it's an agent orchestrator. It it looks a lot like conductor. Uh we've been testing it internally for the last couple weeks and uh it looks like this. We like it. I've switched about 50% of my usage from cloud code to this. Uh Naveen, who's a Codeex guy through and through, has switched all of his work from out of the terminal and into um into this app. And I think that's actually really really interesting. It's a it's a trend that um I would guess we will see more of and you're hearing it today, but I think in like three months it'll like it'll be a little bit more of a mainstream thing which is um terminals have had a resurgence over the last six months u driven by cloud code which they figured out these like really cool patterns where uh if you have an agent on your computer it has access to everything on your computer can do everything that you can do and it has access to the terminal it can write bash and that's like a very very powerful combination and so I think that has created this whole explosion of agentic coding and the codeex app is the first app since cloud code came out that has made me want to do some coding in a guey instead of in the terminal. Um turns out that gueies like have some good stuff. Naveen was saying you know it's really easy to to see to to discover features because you can just go over here and be like oh there's like a there's a little button here that like gives Codex full access over your computer. um you don't have to set a flag. You don't have to do a slash command. Um and there's still going to be terminals around, but it's uh I I think we'll see this more and more. You can also see this with cloud code. They're they're they have a desktop app now. They have cloud co-work. Um so they're sort of starting to move in the same direction, but it's it's a general trend that I think is happening more and more where goes do not feel like a step back in the way that they they have for the last couple months. So we speedrun we speed ran the terminal era in only 3 months or 6 months. \>\> Yeah. Yeah. And uh yeah just to I think I'm really curious where this uh thing is going to go like how this particular UI is going to look in 3 months down the line. Right. I think people are going to use like codeex or cloud code apps more and then you're going to just spend most of your time in this app uh driving everything there's still ton of features I missed right now it's just the initial version right but I'm curious how they do reviews how like you know once the code gets generated what happens post that \>\> so \>\> totally totally um yeah you can you can see that this is the new race that I Claude has Enthropic has been running with Claude for the last like six months and I think codeex just like dipped their toe into entering which is if you build a good enough programming a programming agent on a person's computer it's actually really good for any other kind of work that they want to do. Um I think that's where that's where Claude shines. That's what Anthropic has figured out is it's a friendly fast model that people love to work with that started being really useful for programmers and now like for example people on our team are growth people are just doing five hours of work in 30 minutes with it and uh designers the same thing and marketers the same thing and so um I think anthropic is pushing pushing that boundary. I think previously codeex has really been marketed for senior engineers and I think it's still actually targeted at senior engineers or I at open eyes marketing has shifted to positioning it less for senior engineers and more for everyone but the reality of the app as it is today is that it is still actually \>\> um mostly for people who are doing this the kind of like high-tech um really high performance coding that Naveen is doing but they're moving in that direction and I I would guess that they'll they'll invest more in that over time And it'll be interesting to see what they do with the branding. Um whether they keep Codex or do something else because I think the chat GBT is like their more consumey get get work done product. That's \>\> yeah and how that interacts with codec is going to be interesting \>\> 100%. And I think recently they also added this apps feature under experimental where you can just have direct connect to notion connect to like you know all the third party apps. So they for sure uh like want to go into this you know mainstream work but still CEX may be target so far my understanding is coreex is like target towards engineers like you know senior engineers. \>\> Yeah \>\> not like cloud code. \>\> Yeah. \>\> Yeah. Um one thing uh Orion asks who made the animation on the vibe check page? Let's just all like take a second and admire it. Isn't that beautiful? Um, and in particular, one of the details that I love, and I don't even know if this is intentional, but it says gold here, and the daffodils that are in like there's coded daffodils. The daffodils that are in code or aski art or whatever, um, are also gold and that's like really fun. The person who did this um, let's see if the person did this is our creative lead or actually he's creative director now. Um, Lucas Cresbow. Let me open that up. Uh, Lucas is incredible. He's such a great designer. He does everything at every um he does not love the Granola rebrand. Sorry, Granola. Um and he he basically did this. He does cool like this. Like look at that. Look at that. Watch. Um so he's just like one of those multi-talented people who he's a designer. He he doesn't know how to code, but he's using cloud code all the time and building stuff and he loves using Framer for this actually. So you should definitely follow Lucas if you have not. Let me just check the my tweets real quick and see if there's anything interesting. Um, thank you Joe. Yes, we are on top of things fast. And Colleen, I do think that now that hydration is becoming a thing, um, that we should probably make water bottles or something because it's just super super fun. Um, \>\> every icon on the water bottle. It would be awesome. \>\> That would be really good. That would be really good. Um, speaking of which, I haven't had water in a little while. So, if you're at the edge of AI like us, um things are moving really fast and it's very important to stay hydrated. This is my second glass of water on this stream. I have an entire setup here for the rest of the day, so I can stay hydrated. Naveen does as well. Naveen, show us your water bottle. \>\> I have I have a 1.5 liter water bottle. \>\> So, 1.5 liters, that's a lot. That's a really That's a That's a good amount for you and a good amount for your agent. Make sure your agent is liquid cooled. I don't know what that was, but I finished the glass. So, that's really good. We'll keep going with the rest of the the last rest of the live stream. I'll have to pee at some point, but um yep, just take care of yourself, take care of your body, take care of your agent. Let's keep going. Um what else? What else is there? If you if you're in the chat, we've got about 2,000 people here. Feel free to ask questions. Um, I' we'd love to tell you our experiences with the Codex app. Again, we've been we've been testing it for about a week now. Um, maybe like a week and a half. Uh, I think we both really like it. Obviously, there are some there are some rough edges. There are some things that need to be better about it, but we're spending a lot of our time in it and it's a really great um it's a really great alternative to the terminal for me. Um, anything on your mind that we haven't talked about that you think people should know? Um another interesting thing is if you go to skills tab I think skills skills is something you know it's like you know difficult for you to see what's important like you know what should I do and then they have this like recommended uh skills that's really interesting uh they have like Figma atlas like so they started \>\> oh they have an atlas skill oh my god let's just try the atlas skill um \>\> oh we should do that and then linear like I'm curious about the linear because linear is my project management level. So, uh it's \>\> I don't use linear, but uh but you should maybe you install it and we could try your the linear skill and then I'm going to try the uh that atlas skill. So, let's just say I'm going to say like can you use Atlas to check our chart mobile and tell me our current MR. I'm just going to like do the most basic thing possible. \[snorts\] Um I probably should not have used GBT 5.2 Codex high. But basically like this is a good demo if you're just joining. We've got a whole uh Codeex app here. On the left is uh orchestrators. It's it's using it's using Atlas. It's it's off the screen right now, but it's it's definitely using it. And that's kind of cool. \>\> Um \>\> interest. It's not a headless browser. It's like it's on Atlas. One of my like little the little quirks of it is that what the Okay, I can't read the dashboard values now. It only manages tabs. So, I need you to tell me the MR shown. What does that even mean? \[laughter\] \>\> Uh, wait. You can't see what's on the page. \>\> Oh, that's \>\> See, this is one of those things. This is like this is like why um uh this is I think characteristic of the codeex one of the one of the codec annoyances which is it's it's typically very hobbled. Um it's typically like one of those things where it's it's kind of gated in a way that you might not expect but that someone was thinking about because it's like a security concern but it means that the whole thing is like less valuable and that that just it like harshes the vibe a little bit. And I think Anthropic is generally has been a little bit better at um opening things up in a way that if you're if you're like a a really senior engineer, you might be like, "Oh this is I don't know if I want this thing to have access to my computer in this way." But if you're just like trying to check your MR with Atlas and Codeex, you're like why can't this why can't I do this, you know? Um so uh I think they're getting better at this, right? They added this um they added this uh what is this? Look at this. Codeex is full access over your computer. Elevated risk thing that you can turn on and off, which is really great. It used to be a flag, but this this stuff still shows up. I don't know if this is this might not be actually the Codex team's thing. It might be that the Atlas browser itself doesn't allow you to do this, but like it's still uh I think it uh is indicative of of a whole vibe that's probably in a lot of different places inside of OpenAI right now. \[snorts\] \>\> Yeah. Also, maybe they just add it just for fun. Looks like they you can only control the tabs and stuff. Other than that, you can't really control the \>\> That's part of the That's what the skill is. Yeah. \>\> Yeah. \>\> Interesting. \>\> So, interesting. Other interesting things would \>\> have you have you used any automations? \>\> No, not really. Like, uh I want to, but uh again I Yeah. So, it's a different thing, right? Like again you're kind of like driving the whole coding as and then they have this automations \>\> that they just added it but not like I don't know what to use it for. I hope \>\> my idea was like look at all the things you learned all the bugs you fixed and update all the appropriate places like cloud MD agents MD and obviously this is weird weirdly transcribed and project learning documentation with everything that we created with everything we learned to make things better for future agents. So like but it looks like there's no message found so I don't really understand let's just test it. I I don't know if it's actually working for me. Um, that's maybe a that's like a sort of a small little quirk is like how would I know if this worked? Shouldn't I shouldn't I see it? Oh, I guess there's a thread here. Okay, \>\> but it seems like if if I press test, it seems like that is actually working. But, um, yeah, I don't know. Okay, Mr. Golden Rose asks um as someone who excels in system architecture using personas and is starting to design via vibe coding with lovable with success and moving to cursor plus amp. I'm thinking codex might help me QA possibly. It sounds like um I would say if you're using if you're mo moving I assume you mean conductor and amp. Um it sounds like codecs might be an interesting one for you. depends on um how big the projects are and what kind of projects they are. My general take is um if you're working in a big codebase and you care about performance and care about um bugs not not uh showing up, Codex is is quite good. Also, if you're working on like anything to do with math, so you'll see in my codeex over here, I have our every growth OS. This is a a skill with a bunch of connectors that allows us to track things like our MR growth and like um cohort churn and all that kind of stuff. I think codeex is good for that because as Naveen said, it's like that grumpy senior engineer who you kind of hate talking to but is always right. Um and cloud cloud code is more likely to get those things wrong. And for anything with numbers, it's just like maybe Codeex is better. The trade-off is just a lot slower. So, if you're interested in that, I think it's um it can be good. If you're interested specifically for QA, I mean, I would have it fix bugs and write tests for me, but based on what I'm seeing right now, I think just claude with its browser is going to be a better better option. Um, at least for now. Um, uh, okay. So there's a follow-up question that says code generated in Gemini which only provides samples most of the time can codeex help and and a half step towards cursor. I honestly don't know what that means but I would love for you to clarify in the chat. I'm just noticing I'm um pretty much out of time here. Uh we've been streaming for about an hour. \>\> Um uh if you're just if you just got here, you should check out our full Vibe check. It's on every we've been testing the Codex app for um about a week and a half week a week now. We have a full Vibe check here. I'm going to throw this in the chat. Um it's every.TOVEC-VIE-check. Um it's all of our thoughts, all of our in-depth testing on what the Codex app is, and um what what we like it for, what we don't like it for. Um you can you can watch the the recording of this full live stream here. You can see the reach test. The reach test is our way of telling you did we reach for this model or this product dayto-day. Uh so most of us are green, one of us is yellow. So a lot of us are using it a lot which is I think a good leading indicator. There's also a full analysis on every. So this is just like the a taste of uh of all the stuff that we've done. We do really in-depth testing and research. Um the full paid article is on every here. I will throw this in the chat. Um, if you are a uh if you're a paid subscriber, you get access to this and you get access to all the other um vibe checks that we do. So, anytime a new model comes out or new product comes out from Anthropic or um OpenAI or uh uh or Google or any of the other big companies, we always get access to it early and we tell you on the day of uh uh whether it's good or not. Um once again just for people who are just joining um we are from every is the only subscription you need to stay at the edge of AI. We do ideas apps and training. On the idea side we do things like these vibe checks and these live streams. On the app side, we have a suite of AI AI apps that we build for ourselves and give to give to you, our subscribers, like Monologue, a smart dictation app, which Naveen builds, Sparkle, which is an AI file organizer, Kora, which is an AI assistant for your email, and Spiral, which is an AI ghostwriter. Um, we also do trainings. So, we have a really cool um Open Claw camp coming up on Friday for paid subscribers. So, if you're if you're watching the the lobster hype and you want to get your uh get your own very own lobster set up, we will show you how to do that and also share with you some uh like learnings from our team like what what we use it for, we don't what we don't use it for. Help you try to separate um the hype from the um the signal the signal from the noise, the mold books from the actual real use cases. and um and we're doing that with Nat Elias who's a friend of every who's really awesome and is like very far ahead in this space. So that's all what you get as part of an every subscription. Check out every.TO if you're not a paid subscriber and subscribe. We also have a discord where you can interact with us and all of our other subscribers. That is all we have for today folks, but there's actually a lot more uh coming up. So, uh, you know, please follow us, uh, subscribe, and, uh, and remember, always stay hydrated. I'm I'm now about to start off on on this bad boy. Um, but that's what you need to do to stay at the edge is stay hydrated. Uh, uh, keep yourself uh keep yourself full of water, keep your agent liquid cooled, and um, thank you, Naveen, for um, for joining us, and for sharing your vibe check with us. Thank you then. And stayated everyone. \[laughter\] \>\> All right. Have a good one. Bye everybody. \>\> See you.