On this episode of the AI & I podcast, Every's Dan Shipper interviews Every's head of growth Austin Tedesco about why he switched from Claude Code to OpenAI's Codex after the GPT-5.5 release; Austin now spends about 80% of his working time in the Codex desktop app. Topics: how Codex went from a tool for senior engineers to a daily driver for knowledge work; how Claude Code proved a great coding agent works for any knowledge work; Austin's Codex setup with folders, keys, and reviewer agents; brainstorming automations across Gmail, Slack, and Notion; managing the human review step when Codex drafts communications; building specialized agents; synthesizing meeting transcripts and Slack threads into a go-to-market plan; building a live KPI tracker in Notion that agents can read; and using Codex for recruiting. The episode argues that the agent management interface—a desktop app built on top of a coding agent—is becoming the new operating system for knowledge work. # Channel Presentation Observations — "Why We Switched From Claude Code to Codex" ## Observable Context & Format - This is a **live-streamed community event**, not a standard produced video. The host opens with: *"Hello, everybody. Welcome to Codex Camp, Codex for knowledge work."* - Format is a **host + guest conversation with live audience Q&A** — audience members introduce themselves and ask questions (Margaret from Plymouth, Alex the musician, Rory, Rich), indicating a real-time, interactive community setting. - The event features a **guest demonstration with screen sharing** — the head of growth walks through actual Codex workflows (automation setup, KPI tracker, recruiting searches), suggesting the channel demonstrates tools rather than only discussing them. ## Explicit Self-Description - The closing promotion states: **"Remember Every is the only subscription you need to stay at the edge of AI."** - The host frames the audience relationship explicitly: > "This is our gift to you as every subscriber. We try to do stuff like this all the time... We've given out... cursor credits... We have more stuff like this coming. So we just want you to be able to try these tools, be at the edge with us." - The event includes a **sponsor read** (Bilt) and a **subscriber-perk giveaway** (OpenAI-provided ChatGPT Pro credits, ~$100 value for ~250 attendees), showing a commercial, subscription-funded model. ## Intended Audience (as demonstrated) - **Knowledge workers and AI-curious professionals**, addressed directly: *"if you're here, you are getting Codex credits"* and audience members self-describe as musicians, agency/leadership folks, and general knowledge workers. - The channel assumes audience familiarity with frontier AI tooling — terms like Claude Code, Codex CLI vs. desktop app, Opus, GPT-5.5, Cursor, and agent workflows are used without basic explanation. - The host repeatedly frames advice toward tool adopters: *"for people who get all the benefits of being able to use these tools, it's really important to be bouncing around between these."* ## Claimed Scope / Subject Matter - **AI tools for knowledge work** is the core demonstrated scope — the host frames a thesis: *"If it can write software on its own, it can do any kind of knowledge work on its own."* - **Comparative tool evaluation** across vendors (Anthropic, OpenAI, xAI/Cursor, Google) is a recurring framing: *"that's the race that is happening."* - **Internal company practice sharing** — the hosts discuss Every's internal workflows (Think Week, Compound Engineering practices, plus ones, KPI tracking), positioning the channel as a window into a working AI-forward media company. - The channel promotes its **own products as part of the ecosystem**: Spiral, Monologue ("our speech to text app"), Cora, plus ones, proof docs, and a public GitHub fork ("compound knowledge"). ## Creator Claims vs. Demonstrated Patterns - **Claim:** Codex is now the speaker's daily driver — *"Codex is the first thing I open... It's where I spend like 80% of my time working."* (Self-reported usage, not independently verifiable in the material.) - **Demonstrated:** Concrete workflows are shown via screen share — automation of Slack/Gmail/Notion drafts, a Notion KPI database rebuilt weekly, a recruiting search for General Assembly alumni filtered by AI experience, and a six-agent Slack suite. - **Opinionated editorial stance:** Strong subjective judgments are stated directly, e.g., *"Codex is one of those things where 3 months ago, 6 months ago, it was trash"* and *"nothing has ever made me feel more stupid than Codex like 2 months ago"* — indicating the channel presents candid, personal evaluations of AI products. ## Audience Engagement & Growth Behavior - The host makes a direct growth ask: *"We would love it if today you would go tell one of your friends to go subscribe to Every."* - Follow-up email and chat distribution of resources (prompts, the compound knowledge repo, credit link) indicates a **newsletter/community funnel** attached to the video content. - The video ends with what appears to be an **appended outro from another show** ("smash that like button and subscribe to AI and I... Dan Shipper as the captain"), suggesting cross-promotion or repurposed footage — no channel-wide identity conclusion can be drawn from this segment alone. ## Anomaly Note - A mid-transcript ad break ("An idea by Avery") and the trailing outro's tone and channel name ("AI and I") differ from the live event framing; the host's name "Dan" and "Dan Shipper" references are consistent, but the source material itself does not state the publishing channel definitively. # Delivery Observation: "Why We Switched From Claude Code to Codex" ## Format Overview - **Two-host conversational structure**: The session is primarily a dialogue between Dan (host/interviewer) and Austin (head of growth, demo presenter), framed as "Codex Camp" — a live subscriber event with audience Q&A and sponsor/ad reads. - **Segments**: opening hype montage → Dan's framing monologue → Austin's personal switch story → screen-share demo → audience Q&A (Margaret, Alex, Rich, Rory) → credits giveaway → closing call to action. ## Tone - **Enthusiastic and superlative-heavy**: Dan repeatedly uses high-intensity enthusiasm markers: - "Psyched to have you, psyched to have you on this auspicious GPT 5.5 day after release day." - "This is so sick." - "It's amazing." - "It's amazing... Feeling really good." - **Blunt, informal candor from Dan**: He opens with provocative honesty and doubles down on it: - "3 months ago, 6 months ago, it was trash. If anyone from OpenAI is on the call and listening to that, I stand by that 100%." - "it was like a little autistic, like it it didn't have any emotional intelligence." - **Casual, colloquial register** throughout: "vibe coded app," "agent pill moment," "holy light bulb moments," "needle in the haystack," "dumb agent... the dumb ones that just do the right thing every time." ## Vocabulary & Recurring Phrases - **"Knowledge work"** is the dominant repeated term (appears many times: "Codex for knowledge work," "any kind of knowledge work," "knowledge work that requires strategic thinking"). - **"Agent / agent-first / agent management"**: "agent management interface," "agent-first world," "your agent is your interface." - **Quantified time/habit claims**: "80% of my time" (repeated by both speakers), "12 hours a day," "maybe 30 to 40% better," "80 to 90% of the way there." - **"It just works"** and variants: "it just works. Like it just works really quickly and and well," "But mostly it just works." - **"Game-changer" / "step change"**: "such a game-changer for me," "such a step change in how you work." - **Concrete tool/product name-dropping as vocabulary**: Codex, Claude Code, Claude Co-work, Notion, Slack, Gmail, Stripe, Warp, Ashby, PostHog, Cora, Monologue, Open Claw, Cursor, anti-gravity. ## Teaching Habits - **Narrative arc of before/after**: Both speakers structure lessons as past-dissatisfaction → present-satisfaction. Austin: "nothing has ever made me feel more stupid than Codex like 2 months ago" → now "Codex is the first thing I open." - **Live screen-share demo as core teaching method**: Austin narrates workflows in real time ("So, this is the Codex app. I I was going to do like a very, very quick tour") and walks through concrete artifacts (folders, KPI sheet, run-of-show automation). - **Concrete micro-case studies**: Austin's go-to-market plan story ("maybe it'll get like a six out of 10 or a seven out of 10"); Dan's recruiting story ("Can you just get a list of GA alums?... it just gave me a list of people"). - **Explicit recommendations to the audience**: "I highly, highly recommend," "I would start here in like a brainstorming automation state," "normalize sending agent documents around." - **Rephrasing audience questions before answering**: Dan does this twice — "but to say it back to you, like what what I'm hearing..." and "I I also sort of read your question, Rory, and you tell me if this is wrong." ## Communication Patterns - **Heavy filler and self-repair speech**: Frequent "um," "uh," repeated words ("the the the the first thing I felt," "I I I use Codex for everything"). This is consistent across both speakers. - **Quoting imagined dialogue as illustration**: Both speakers reenact their prompts and conversations: "Make the run of show.", "why don't you just do what I'm recommending?", "Oh yeah, we need to do a run of show for this camp." - **Interruption-free turn-taking with enthusiastic affirmations**: Dan punctuates Austin's demo with short validations ("That's brilliant. Thank you.", "Sick.", "I love that.", "Totally totally agree with that"). - **Interactive audience handling**: Dan deliberately breaks format — "Normally we take questions at the end, but I think it would be kind of interesting" — and manages names/hands ("Margaret, welcome," "Alex, please introduce yourself"). - **Ad-libbed handling of live logistics errors**: The credits giveaway goes sideways and the speakers troubleshoot aloud: "Um interesting, not available on my plan... correction, this is only if you do not have a plan." ## Stable Patterns vs. Isolated Moments - **Stable**: superlative praise, before/after framing, concrete tool naming, quantified claims, imagined-dialogue quoting — all recur multiple times across both speakers. - **Isolated/uncertain**: Dan's "it was like a little autistic" remark and the closing scripted-seeming outro ("smash that like button... Dan, I'm absolutely, hopelessly in love with you") appear once each; the latter reads as a produced segment appended to the recording rather than a live delivery pattern. Evidence is insufficient to characterize either as habitual. # Position Record: "Why We Switched From Claude Code to Codex" ## Speakers - **Dan Shipper** — host (identified in transcript closing as show captain), Every - **Austin** — head of growth at Every - **Audience members** — Margaret, Alex, Rich, Rory (self-introduced) --- ## Central Claim: Codex Has Become the Preferred Agent Interface for Knowledge Work ### Dan Shipper's Positions - **Codex was previously poor**: "3 months ago, 6 months ago, it was trash. If anyone from OpenAI is on the call... I stand by that 100%." - **Diagnosis of OpenAI's original strategy**: OpenAI theorized (starting with GPT-5) that "vibe coding" would live in ChatGPT while senior engineers would use a hobbled, sandboxed Codex for pair programming — it "would argue with you, it would make you feel stupid... it didn't have any emotional intelligence." - **Anthropic's insight (as Dan attributes it)**: A fast, smart, emotionally intelligent coding agent with computer access is great for *any* knowledge work: "If it can write software on its own it can do any kind of knowledge work on its own." - **OpenAI pivoted**: Over the last ~3 months, Codex went from a senior-engineer pair-programming tool to his **daily driver** for engineering, writing, and recruiting. - **"New operating system" thesis**: Agent management interfaces (Codex desktop, Claude Code/Co-work) are becoming the core surface for work. Each model company has its own: Anthropic, OpenAI, xAI ("essentially bought Cursor"), and Google (Antigravity — "no one is seriously using it for that yet, but I imagine Google will do this, too"). - **Horse-race framing**: Companies will "keep trading" the lead every few weeks/months; eventually ecosystems will diverge. Migration between them is "not trivial, but it's pretty easy." - **Recruiting use case**: Asked Codex to list General Assembly alums filtered by subsequent AI work; "The first one I clicked on... was perfect... one of those holy light bulb moments." - **AI-written documents**: "Normalize sending agent documents around... I would actually prefer to read your agent's writing than your writing in a lot of cases." The test: "do you stand by it? Have you thought about it?" - **Culture argument**: Tools change so fast that "someone using a new tool with a new paradigm... is just going to beat you by default." Advocates play time (e.g., "Think Week" twice a year). ### Austin's Positions - **Agent pill moment**: A week deep in Claude Code CLI (~Dec–Jan), automating work and personal life — "this is the only way I want to do the kind of knowledge work that requires strategic thinking and data analysis." - **Early Codex criticism (preserved qualification)**: "Nothing has ever made me feel more stupid than Codex like 2 months ago"; its clarifying questions felt like "why don't you just do what I'm recommending?" He stayed for engineering results but ~80% of his work remained Claude Code. - **Current assessment after the new GPT model**: - Model quality: "at the very least there's parity" with the latest Opus for his knowledge work; "a few things that Opus does better... a few things that Codex does better" — Opus still trusted for design. - **Real differentiator is the app**: "there's no comparison for how fast and powerful the Codex desktop app is" vs. the Claude desktop app; "the sub agents are so good." - Codex is now where he spends "like 80% of my time working." - **Migration framing**: Friends reacted with "a look of horror"; he estimates the Codex app is "maybe 30 to 40% better" but acknowledges the Claude desktop app could catch up "any week." - **Concrete workflows demonstrated**: - Connected Gmail/Slack/Notion/Stripe; Codex drafted a run-of-show and pushed it to Notion and Slack unprompted-by-hand. - A daily "dumb agent" automation compiling unanswered messages and drafting replies. - Go-to-market plan: Codex assembled existing meeting transcripts/Slack threads into a plan "basically 80 to 90% of the way there" — framed as context-loading and assembly, not strategy origination. - KPI tracker in Notion: **preserved limitation** — one-shot attempts were "5 to 10% off" on numbers; "our MRR number can't be 5% off," forcing manual column-by-column verification. "It feels so stupid that I have to do this... but it makes sense." - **Final review practice**: Drafts in Codex but does the final human-facing review *outside* the app (Slack drafts, Gmail drafts) — "the last pass before humans engage... step away from this agentic space." - **Safety/filtering approach (to Alex)**: Have the agent *interview you* rather than dictating rules; set "never send anything for me, only draft" rules; schedule 72-hour audits of what the automation archives. - **Suite-of-agents approach**: Inspired by a Claire Boucher ("Claire Bo") interview with Lenny — six specified agents rather than one supercharged agent; they "still break," but Codex can fix their architecture when they do. - **Acknowledged tension (to Rory)**: He over-indexes on "playing around and building" tools at the expense of day-to-day business tasks; Codex automations partially compensate. "I don't really have like an awesome answer for it." - **Compound Engineering**: Uses Kieran Plunkett's plugin; forked a public "compound knowledge" variant for knowledge-work reviews (strategic alignment, data accuracy) since the original reviewers target engineering. Also instructs agents to ask "should we compound this?" after sessions. --- ## Audience Positions - **Margaret**: Asked how the "explicit approval" review step works in practice. - **Alex (musician)**: Concerned about email safety — not losing leads or "send[ing] something that loses me money." - **Rich**: Asked whether Compound Engineering is used off-the-shelf or forked. - **Rory**: Asked what team practices help make time for these workflows amid busy schedules. --- ## Promotional / Non-Position Content - **Sponsor reads**: Bilt (rent/mortgage rewards), Every/Avery subscription, "AI and I" subscribe appeal — recorded but not substantive claims about the video's topic. - **Credits offer**: 250 attendees receive a free month of ChatGPT Pro Lite (~$100 value), new users only; presenters acknowledged on-air that existing subscribers were excluded and promised to "try to get something." ## Unresolved Tensions (Preserved) - Codex's model vs. app: both speakers attribute the switch primarily to the **desktop app quality**, not clear model superiority — model quality is framed as parity, with per-task differences. - Austin simultaneously praises Codex automations ("they just work incredibly well") and notes persistent fragility (agents "are going to break a bit"; KPI numbers repeatedly 5–10% off; rule-following degrades with large context). - Strong endorsements coexist with hedged predictions: Claude "could ship versions where it's faster and better" any week. # Speaker & Provenance Observation Record **Video:** Why We Switched From Claude Code to Codex (id: 01a0f15a-87c6-757c-91fb-c7f95c0232e3) ## Speakers & Roles - **Dan Shipper** — **Host / creator** (Every). Leads "Codex Camp," interviews Austin, answers audience questions, delivers closing promotional call-to-action for Every subscriptions. - **Austin** — **Guest**; identified in-transcript as **Every's head of growth**. Demonstrates Codex workflows, shares screen, answers audience questions. - **Margaret** (audience, Plymouth) — **Q&A participant**; asks about review/approval steps. - **Alex** (audience, musician) — **Q&A participant**; asks about email safety and filtering. - **Rich** (audience) — **Q&A participant**; asks about Compound Engineering plugin usage. - **Rory** (audience) — **Q&A participant**; asks about time management practices. - **Unnamed narrator voice(s)** — deliver intro/outro ad copy and the closing "AI and I" subscribe pitch; role: **narrator/promotional reader**. ## Referenced Third-Party Sources & Claims - **OpenAI / Codex / ChatGPT / GPT-5.5** — Dan claims Codex "was trash" 3–6 months ago and describes an alleged OpenAI product strategy (vibe coding in ChatGPT, hobbled Codex for senior engineers); explicitly notes he has **no inside information** on Anthropic's roadmap. - **Anthropic / Claude Code / Claude desktop app / Opus** — frequent comparative claims by both speakers. - **xAI** — claimed to have "essentially bought Cursor" (third-party claim, speaker assertion). - **Google (Antigravity)** — speaker claim that no one seriously uses it for agent management yet. - **Claire Vo interview with Lenny** — cited by Austin as inspiration for his suite-of-six agents approach. - **Rachel Karten's Substack newsletter** (social media frustrations piece) — cited by Austin. - **Kieran Plunkett / Trevin** — credited as builders of the public "Compound Engineering" plugin; Austin's fork "Compound Knowledge" described as publicly on Every's GitHub. - **Built/Bilt ad** — sponsored segment read describing rent/mortgage rewards, United, Hyatt, Lyft, Amazon partners. ## Promotional Segments (flagged) - **Bilt sponsor read** mid-transcript ("Make sure to use our URL so they know we sent you"). - **Every/Avery promotional narration** ("the only subscription you need to stay at the edge of AI"). - **OpenAI Codex credits giveaway** — 250 attendees, ~$100 value ChatGPT Pro Lights month; noted limitation: **new users only**, existing-plan users excluded pending alternative. - **Closing Every subscription push** ("tell one of your friends to subscribe"). - **"AI and I" channel outro** — like/subscribe pitch naming Dan Shipper. ## Provenance & Availability Notes - **Internal/own-company tools referenced:** Plus Ones (hosted Open Claw), Cora (AI email assistant, run by Kieran), Monologue (speech-to-text app), Spiral, Proof (markdown doc tool), Notion KPI tracker, Doist reminders, Warp CLI, PostHog. - **Named third parties mentioned:** Kate (editor-in-chief), Brandon (COO), Jason (audience interjection), General Assembly (recruiting source). - **Comment availability:** not stated in transcript or supplied metadata. - **Uncertainty:** speaker identities inferred from in-transcript introductions ("I'm here with our head of growth, Austin") and the outro naming Dan Shipper; audience Q&A timestamps not provided in transcript. Codex is one of those things where 3 months ago, 6 months ago, it was trash. If anyone from OpenAI is on the call and listening to that, I stand by that 100%. If you have a great general purpose coding agent on your computer, it's actually really great for any kind of knowledge work. If it can write software on its own, it can do any kind of knowledge work on its own. \>\> When I sign on during the day, Codex is the first thing I open. It is pulling in whatever I need from Gmail, Slack, \[music\] Notion, Stripe, all of our data sources. It's where I spend like 80% of my time working. Overwhelmingly because the app itself is just so good. \>\> There's a new operating system for how and where you're going to get your work done, and it's this kind of agent management interface. \>\> \[music\] \>\> Hello, everybody. Welcome to Codex Camp, Codex for knowledge work. Psyched to have you, psyched to have you on this auspicious GPT 5.5 day after release day. Hope you're doing well. Um I'm here with our head of growth, Austin. Austin, say hello. \>\> Hello. \>\> We're psyched to have you. We are psyched to do this. Um Codex is one of those things where, you know, 3 months ago, 6 months ago, it was trash. Um and if anyone from OpenAI is on the call and listening to that, I stand by that 100%. Um and it was really built for um senior engineers uh doing pair programming. So, it was um it would argue with you, it would make you feel stupid, it was just it was like a little autistic, like it it didn't have any emotional intelligence. And I think OpenAI had this interesting strategy or this interesting theory starting with GPT-5 that your vibe coding was going to happen in ChatGPT and that was where all that stuff was going to live and then senior engineers are going to use Codex to like do all of their programming work but we're going to hobble the model so it doesn't do anything bad as in a sandbox all that kind of stuff. And I think basically what happened is Anthropic figured out that having a model that's pretty usable and fast and smart and also emotionally intelligent intelligent on your computer that can access your computer um is a really really great experience for programmers and it means you could throw away a lot of the old uh stuff that you used to have in a in a programming environment where you were it was built for typing code you could just type commands into your terminal and then it would start working. And then I think what Anthropic figured out is if you have a great general purpose if you have a great coding agent on your computer it's actually really great for any kind of knowledge work. If it can write software on its own it can do any kind of knowledge work on its own. And we started to move from this world where programmers had been delegating um had been delegating their tasks starting to delegate their tasks inside of Cloud Code to now any kind of knowledge work is being delegated inside of Cloud Code and Cloud Code work and all that kind of stuff. And I think OpenAI they had this original split it's like oh you're going to do all your vibe coding in ChatGPT and I think they saw what was starting to happen with Cloud Code and over the last maybe 3 months or so they have done this hard pivot on code Codex where it has gone from a senior engineer only tool that it's really for pair programming um to I think like it's it is my daily driver for this kind of work. Um I uh I I use Codex for everything from deep engineering stuff to writing to recruiting I do a lot of actually do a fair amount of recruiting it's really good for that. I'll give you some use cases um later. But they sort \[snorts\] of figured out that um having this general-purpose agent on your computer with the ability to write code, the ability to access your file system, the ability to have a browser, um and wrapping it in a desktop app is like the ideal uh ideal next step for knowledge work, and I think that they built the best current version of that. Um and it what it is starting to snap into into focus now is that there's a new operating system for how and where surface for how and where you're going to get your work done, and it's this kind of agent management interface. And that's whether or not you're using Claude Code or Claude Co-work in the desktop app or Codex in the desktop app, it's becoming this race between the model companies where every each model company has their own surface like this for agent management, a desktop app for agent management that's at its core a programming agent that's used for knowledge work. Um Anthropic has Claude Code and Claude Co-work. OpenAI has Codex. XAI recently um essentially bought Cursor. Um and uh and Google is the only one that I mean, they have anti-gravity, but I don't think no one is seriously using it for that yet, but I I imagine Google will do this, too. And that's the race. That is the race that's happening. And so, I think for us who gets who get all the benefits of uh being able to use these tools, uh it's really important to uh be be bouncing around between these. So, like use it for example, using Codex so that you can feel what it's like to work in an agent-first world because once you add once you add an agent that is like that your primary way of accessing and using software and the internet and all that kind of stuff, it opens up all this interesting stuff that wasn't possible before because you can send your agent out to go talk to other pieces of software and come back and um you know, we can get into into more of the details there, but I want want get into like the more of the concrete use cases, but that's the world that we're starting to live in. You're doing work on your computer through Codex or Co-work. And um and your agent is your interface to a lot of the work that you're doing and a lot of the a lot of the software that you use and a lot of the stuff that you do every day. And uh that's actually really fun. It's really cool. There's a lot of good stuff here. And so I wanted to uh I wanted to bring Austin in to to help do this because Austin is our head of growth. And I think he had his real agent pill moment. You tell me, Austin, but probably like three or four months ago. And the agent pill moment was really Claude Code. And I sort of remember you just being like "Oh, yeah." On a on a Monday morning being like, "Oh, yeah. I just was on my computer all weekend. Like I I was like 12 hours a day. I didn't go out or anything." Cuz I was using Claude Code. And um you started to use it for all those all the kind of knowledge work tasks that a that a growth marketer would. And over the last couple weeks as we've been using 55 and I've been telling you for a little bit you should try Codex. It seems like you've you actually just shifted everything over to Codex in 55. And so I think you're a great person to talk about, you know, sort of what you're seeing and how and how that is uh how that is um how this has changed how these agent management interfaces have changed your workflow. And then why you like Codex. And then I would love to get into some demos of your actual Codex workflows so that we can sort of see things uh from your perspective. \>\> Yeah, that sounds great. So I um yes, my like agent pill moment was spending a week going deep into Claude Code in the CLI. Uh probably in like December into January hooking it up to everything I do for work and for my personal life and finding that I I use Warp as my like CLI interface and finding that the things that could automate the things that could handle for me and then the way it could work as a thought partner to make my work better. It was like this is the only way I want to do the kind of knowledge work that requires strategic thinking and data analysis and shipping marketing copy like a bunch of stuff that can get you spread out across of apps and tools during the day. And and maybe in February you kept nudging me to be like you really should try Codex. There were things you liked about it and and if someone says that it ever if anyone on the team says that like I'll go try it. And I like to push myself and play around with more engineering tasks especially to see what these models are capable of. And so I tried to build a personal vibe coded app in Codex cuz that was one of the things that you said that it was really good for. And my immediate response was like I think it is better at building the app but I can't tell because it's nothing has ever made me feel more stupid than Codex like 2 months ago. Like I always I used Compound it the Compound Engineering plugin that Kieran Plunkett made for basically everything including knowledge work but especially if I'm trying to build an app or ship a PR to the to the site. So I made a plan in the plan it comes up with like three questions and for like which direction we should go and I had no idea what the hell it was talking about. It was like do you do one any of these three and every question I was like please explain this to me in more detail and its response was basically like why? Like why don't you just do what I'm recommending? And I found a way to I I basically stayed in Codex for all engineering stuff cuz I did like the results even if I didn't love working in it but I would say 80% of what I was reaching for was was Cloud Code in the CLI. And when we got our hands on the new GPT model a a month ago, the the the the first thing I felt was at the very least there's parity between the latest Opus model and the latest GPT model for the kind of knowledge work I do. There's a few things that Opus does better. There's a few things that that Codex does better. That feels a little more specific to me even. Like I I outside of design, which I still really trust Opus for. Um it feels a little more like a there's some stuff I like better than this. But the real differentiator to me is that to me there's no comparison for how fast and powerful the Codex desktop app is as just like an app compared to the Claude desktop app. Like I have never been able to get uh code work to work for me. And I think it's because I've been kind of ruined by the Codex app. It's so fast. The sub agents are so good. The way in which it suggests and then um ships automations for me is just like I can't imagine not using it. I wouldn't be surprised if any week the Claude desktop app is like just as good, right? Like um they could ship versions where it's faster and better. But I I'm now at the point where when I sign on during the day, I Codex is the first thing I open. It is pulling in whatever I need from Gmail, Slack, Notion, Stripe, all of our data sources. This morning I was like, "Oh yeah, we need to do a run of show for this camp." I messaged Codex. I'm like, "Make the run of show." It knows exactly where to look cuz we've already had conversations about what we're going to talk about today. It pushed it to Notion. It sent it to Slack. It was perfect. It was like, "Oh yeah, this is exactly what we should do." And um yeah, it's where I spend like 80% of my time working overwhelmingly because the app itself is just so good. And then the model has now gotten good enough to be the daily driver for me. \>\> Yeah. And I I feel I feel the same way. Um I love to get into and and for people someone who someone asked, are you we discussing the app or the CLI? We're discussing the app, the desktop app. And and I think you're making a good point that uh both of these companies I think sort of see the end game here and they're pushing in the right direction. And for a while at least, it's going to be a horse race where every couple every couple weeks or every couple months, like one is going to pull ahead and have this like sort of amazing thing going on and then their the competitor is going to uh like Anthropic for example, I think will release something in a couple weeks or couple months. I don't have any inside information into this, but that will make it at least parity, if not better, and they're just going to keep trading. Um and at some point I think that'll slow down and you'll end up with sort of separate ecosystems, but for now, they're actually fairly easy to switch between. It's not it's not trivial, but it's pretty easy. Like you can kind of ask Codex, hey, can you go grab all my Claude stuff and it'll go do it. \>\> It I think it it feels that way when you do it. It's funny. I'm in I'm in New York right now. I usually live in LA. Most of my friends who are in the knowledge work space have been asking me about like what they should be using. They're all Claude code or Claude desktop app filled and when I tell them that I have fully transitioned to Codex, this like look of horror shows up on their face and they're like do I They're kind of like, do I really have to? And I of course tell them they don't, but I'm like, you really should. Right now, you really should. Like I think you would get a big benefit from it and I've been showing them why. And it's it's interesting and and to me unsurprisingly how resistant people have been to it because the when you're a knowledge worker and you have all these new tools, the Claude desktop app is is game changing. It's amazing, right? So the idea that the Codex app is maybe 30 to 40% better is like, "Ooh, that's a lot of work." Um which we can get into kind of how I migrated. I can show some of that. It was very easy and the ways that I'm uh starting to use it. So, I'm happy to dive into that and start sharing my screen and show what it \>\> Yeah, why don't you share your screen? Um I think yeah, I I kind of agree with you. It's more of like an emotional thing of like, "Oh, I have to learn a whole new thing." or whatever, but it's it's pretty similar. Yeah, I would love to see some of your workflows. \>\> Cool. So, um this is the Codex app. I I was going to do like a very, very quick tour. I think a lot of the audience has seen it, but kind of like where I go and how I use it. Um one thing I love about the Codex app is like I do think it's much better organized than the Claude desktop app. My the ability to have these folders with persistent, consistent chats inside of it that I can go check out. And then especially like the big differentiator is that because I do think this is much better for engineering, for like occasionally I will ship a PR for one of our products, it's great to not have to switch between uh the Claude Code CLI, the Claude desktop app, and Codex that I can be here, I can be working on or improving our KPI sheet, which I'll like show what I was doing here, and then I can go down to plus one and ship a PR for plus one. And um the other thing I found cuz I did I tried the new I tried the updates in the Claude desktop app last week when they when they shipped it. And the the stress test I put on it was make a go-to-market plan for our new product and ship a PR to sparkle in different chats. And it it was so clunky and slow. And when you do stuff like that inside of Codex, it just works. Like it just works really quickly and and well. And and and that's the thing that like once you start feeling that, it's very hard for me to turn away from it. So, I have these different folders for um some like five coded apps that I play around with for fun, for my personal open claw where I can go and manipulate it here, and then the one with all of the chats is this like every growth OS. All this is is a folder with a bunch of um secrets and keys, so it's connected to everything we use for every and then some project instructional files that explain what the every business is, what we care about, how we like to work together. It has some like reviewer agents inside of it that are all informed by how Compound Engineering works. Inside of Compound Engineering Kieran's plugin, there is a Compound Engineering review step once you do some work that reviews for like security and a few different things that are often times not as helpful for like I'm doing a strategic plan for a go to market initiative. And so inside of here there's like a fork for it for strategic alignment with the company goals, for data data accuracy, and having that inside of this folder means that as I'm making plans, I can get reviews from the model in like a targeted way. And so the first thing I wanted to show is like how I I was talking to our our editor-in-chief Kate yesterday to show her like how I would recommend getting started in Codex. And this is my recommended prompts. I'm happy to put it in the chat for people. We can put it in the email as well. And so all I did was I'm putting in the prompts here. I only have personal panelist access, so I'll send it later or something. \>\> Okay, yeah, maybe read it out. \>\> You can all agree that housing is expensive. Rent or mortgage doesn't matter what you're paying, it stings \[music\] every month. But Bilt can make it feel a little better. Bilt started out by rewarding members \[music\] for their rent. Now as of 2026, Bilt members can also earn points on mortgage payments wherever \[music\] they live. Every housing payment earns points you can use toward flights with top travel partners like United and Hyatt, Lyft \[music\] rides, amazon.com purchases, and so much more. This is actually pretty cool and I I some friends that use \[music\] this and like a lot. Something that's underrated is that Built members also get access to a neighborhood concierge. They can make restaurant reservations, book fitness classes, and find new local spots, \[music\] all while still rewarding you at 45,000 merchant partners. It's like having a personal assistant baked into where you live. \[music\] It's simple. Being a renter and now owning a home is better with Built. \>\> \[music\] \>\> Make sure to use our URL so they know we sent you. And now, back to the episode. \>\> Yeah, yeah, yeah. Um okay, I can zoom in as well, I think. There we go. Um so, through the plug-in tool with Codex, I have went in and man and like did the manual clicks to connect all of the tools I use every day, like Gmail, Slack, Notion. And then, I went to a new chat inside of this folder that was built through Claude code. Claude code built this whole Everycode OS system. There's a Claude MD file in there, and it's saved locally. It's also um synced and then pushed to to GitHub. And so, I just opened that project inside of Codex when I started working here. And I uh start a compound engineering brainstorm workflow, cuz it is again just kind of like a thing I reach for of let let's think about this thing together, me and the me and the model. And basically what I said is like, go take a look at the things I use the most, which are Notion, Slack, and Gmail. And think of some automations that would help me with my work. I I find that when I'm trying something new, whether it's a model or an app, um having an agent, having a very smart frontier model like tell me how to use it, tell me what it should do is like exactly where I want to start rather than thinking of it myself. I usually start here. Sometimes I will describe a very specific problem, but this is very helpful for me and I think a good generic problem for people to start with. And uh Codex comes back. It looks at what's going on for me and for the company right now. And um I thought these were really good that like um it has this kind of follow-up radar. This is a big thing that happens with people who do knowledge work, who do partnerships, who do social media marketing that there's all this stuff coming at you across a bunch of different sources. Like what if it handled the triage for you? Um what if it had this kind of like command sensor when we run a a camp or an event which usually requires a bunch of moving pieces and and moving parts. Uh like they mentioned for recruiting and hiring, we don't use a tool like Ashby or something. We kind of have it all synced through Notion because uh apps like this and agents can kind of like handle a lot of the pipeline and tracking work for us and you can just ask it to automate it for you. And so it it does that and it asks me like which ones look good, what do you want to tweak? For the for the sake of this demo, I didn't give it any real feedback. I was like looks good. And um this is actually the thing I've always been most impressed with Codex for and and for the models is that it's like great, I made this automation for you. And I do find that they just work incredibly well. They require very little tweaking to be like this is a thing I would and do use every day. Uh there's this set of instructions that it comes up with based on what it knows about me. I can change when it runs. I can give it additional insights. I can connect it to other things. But mostly it just works. There's there's one that works for me that just at the end of each day now compiles all of the stuff that I haven't responded to yet, drafts the replies, and we can kind of like knock it out together of what to say or like actually all I need to do is just give like a thumbs up Slack reaction or something and it'll do that for me. It's kind of like a like a dumb agent. Like I think of agents like this is like the dumb ones that just do the right thing every time. And then the smart ones like an open claw or a a one of the projects we have coming that's like you'll work back and forth with it and like have a have like a more of like a creative strategic partner. And Codex is good at building both and I can I can show kind of like the smart agent setup, but if someone is looking to be like, "Can I see what this thing can do to help me with knowledge work?" I would start here in like a brainstorming automation state because it is and I think you'll also be surprised by how fast it is and you're like, "Oh, I'm starting to get what this thing can do." \>\> This is so sick. I your your Codex is just far surpassing mine in terms of interestingness. \>\> \[laughter\] \>\> I'm getting a lot of ideas. Um, I want to just actually pause here. Normally we take questions at the end, but I think it would be kind of interesting. If you have a question about what Austin has just showed, it would be nice to let people come up and um just ask a question or two just to see what the vibe of the room is like. Um, so please raise your hand uh if you have a question and we will uh call on you. Uh Margaret, welcome. Uh please uh introduce yourself and ask your question. \>\> Hi. Can you hear me? I'm Margaret. I'm in Plymouth. Um and my question is what does your review step look like? So, it's um saying don't send, post, archive, or modify without explicit approval. So, what does that look like? Is that like do you call up say, "Hey, let's do the review flow now?" Or does it do we push notifications to your phone or what? Thanks. \>\> Yeah, so um for this, what I prefer and I was actually talking to a a friend at dinner last night who said they did the same thing on their own. They came up with this too is like everything I I work primarily in Codex. I do all the drafting and setup in Codex and then it's helpful for my brain to have the final review step actually live in the external app. So, it will draft all the Slack messages and then I can go to Slack where Slack has that like draft reply um tab, and I can go and knock them out. And I do find that it it like uh freshens up my my brain a bit to be like, "Here's where I'll just make sure that this is what I want to send to a human being." Uh same thing for email. It like creates all of these drafts in uh in Gmail, and I'll actually go open Gmail and look at them and and knock them out. I I know some other people who just have it actually come up inside of Codex, and they're like, "Yeah, sure, send it. It looks good there." Um I do the same thing for strategic planning. It pushes to a either proof doc uh the the like agent-friendly markdown file that Dan made or a Notion doc. I use them for two different things. And I I just like for like the last pass before humans engage with it to step away from this um agentic space and have a final check in another surface. That's really the only time that I'm like leaving the app to do something. \>\> That's brilliant. Thank you. \>\> Sweet. All right, we'll do one more, and then we'll keep going. Alex, please uh introduce yourself and ask your question. \>\> Hi. Uh my name is Alex. I'm a musician, and uh I I do a lot of gigs and get uh emails from clients all the time. Uh so I have to sort my leads from, you know, my newsletters and all that informational stuff. So how do you make sure that you prompt uh Codex to um keep those emails safe for me, the ones that, you know, that require a personalized response, and and um I just want to make sure that, you know, I don't send something that, you know, loses me money hehe or something. \>\> Yeah. Um so for me personally, I rely a lot on Cora, the internal like the the app that uh Kieran runs at every for like the AI email assistant that's a part of the every subscription. It's it's really helpful now that inside of Cora there is a like CLI and an API connector that I can work in Codex and tell it tell Cora which is managing my email filtering and my email rules what I want and what I value. The way I do that is the same thing I would recommend whether you use Cora or not which is you have the agent interview you to get an understanding of what the rules should be. I always find that I get a better result rather than saying what I think the rules should be and so I will I'll do a brain dump using monologue our speech to text app saying here's the problem I'm facing my email's a mess let's figure out how to triage it. I think it would work perfectly well if you wanted to try starting it as an automation in in Codex or a rule in Codex of saying like I think these are the things I want to make sure I get. I think these are the rules I want to set of like never send anything for me only draft. I think I want to go through all emails at 3:00 p.m. on on weekdays but go take a look at all my email go do a search spawn sub agents to do a search. I'm always telling Codex to spawn sub agents to do different types of work across different workflows. And then come back with a plan. Come back with a plan for like how you're going to set up my my email. And then you can read the plan and see oh it looks like it's actually going to brief or summarize or auto archive something that might lead to making money for you and that's where you can tweak it. And then the other step I take is that I set reminders for myself in I used to doist for all of my like reminder task tracking. It's also connected to Codex so I can just message Codex or message by Open Claw and say like just add this reminder to my schedule to to check how the new automation is working and like do an audit of it. So you can see like it's been 72 hours. Let's see if I'm see if I've missed anything. You can prompt the model to see like what have you been archiving? And I find that really helpful, but I I'm really excited by all of the work our product and GMs and everyone have been doing to make it so that I can just prompt Codex or Claude Code or you know, inside of cursor any agents to manipulate those apps how I want. It works really well with with our other tools also. \>\> Thanks Alex. I will I will add to that like one of the things that we found basically because Austin started doing it, I sort of was like oh, that's really interesting is that Austin started we have plus ones which is our hosted Open Claw. And Austin started setting up his plus ones with Codex and Claude Code and realized that it's just a much better experience. So rather than for example the earlier version of plus ones, we had like a whole dashboard and a whole onboarding experience where you had to kind of manually click a bunch of buttons and give it a lot of context. It's much easier if we just expose plus ones via a CLI to Codex or Claude Code and then you can just like talk to Codex and it will take everything it knows about you from your computer and your past conversations and throw it into throw it into a plus one setup and and Austin's showing this and it's it's like a it's really powerful and it's it's part of what I'm saying about how the world is changing when you assume every user has access to an agent like this because we don't have to have a settings dashboard. We don't have to have an onboarding experience. We don't have to gather as much context manually. It can just be given to us for free by Codex, and that's really interesting. \>\> Yeah, one of my favorite use cases was I got I got really inspired by this interview with Claire Bo did with Lenny where she said how much of a breakthrough she had when she stopped trying to just use an individual open claw as like a master supercharged open claw and had this suite of six like specified open claws. I think that applies to any kind of like agent like there's the new uh chat GPT like provisional agents like I I got hooked on that and I think Claire's point was really good. And my path towards making this suite of agents to help with the growth function at Every was just going to Codex, going to this folder. Um, I I actually just sent it the transcript of Claire's interview with Lenny and said like, "I want to do this too. Given everything you know about me and my work, um, make a plan to suggest six agents that we should provision into our Slack. Um, consider the fact that we might want to make some of them notion custom agents which I find work really well. It's just like do the same thing every day every time. Some of them might need to be smarter automations, but like do that, come up with the plan." The plan it made was really good and like I tweaked it a bit um after seeing it, but there's that and then now I have this suite of six agents in our in our Slack that that work really well for me. They still break. Like I I find when you're making open claws and and personal agents right now like they're going to you should accept they're going to break a bit, but the really powerful thing is that rather than going back and forth with the agent forgetting frustrated, um, I just go to Codex and I'm like, "I I either screenshot or I can at Slack in Codex and say like, go find this conversation where this stupid thing happened and fix it." And it it does a really good job of just like changing the architecture of the agent and making a fix from there. I love that. Um, yeah, it's it's just a it's such a it's such a step change in how you work and now now I want to paste that clear interview, too. \>\> \[laughter\] \>\> I want to I want to show one I want to show one thing that I like this is like kind of actually my favorite way to use this stuff for for for knowledge work. It's a thing that I like wish I had for so much of my career cuz this is one of the most like time consuming to me like frustrating things about about knowledge work is that we are doing a a real go-to-market market public launch for a plus one we're very excited by it and we've been having a bunch of internal meetings and slack conversations around like how are we taking this to market? What is the strategy? What are we going to do? Um, and we've done all of the work that like kind of like only humans can do the like marketing case the business case the like the narratives and stuff. Not all of it is as refined as it needs to be cuz it still needs to be refined but it's all sitting somewhere and I had all these plans this week to make the go-to-market plan which is like one thing I'm responsible for and an inevitable thing that happens that happens in everyone's job is like all this stuff came up like I've got to do interviews for hiring. We found out the release date for the new chat GPT model and so I had a day I think it was Tuesday in between meetings where I'm just kind of like I'm prompting Codex this way hey I I kind of done most of the work, right? Like in our notion every meeting is recorded in a single place that all the transcripts are there. We've we talked about this a bunch in slack. I have a template for a go-to-market plan that I really like and I can go to Codex and say like can you just make the plan? Like and in my head what I'm thinking is like maybe it'll get like a six out of 10 or a seven out of 10 and we can keep nudging it and I can keep like going along and so it it does that. What I'm asking for is like, why don't you start by doing the compound engineering brainstorm step to to just ship a proof doc and I can see how close you are. And um I one thing that it doesn't really do super well unless I tell it to and I want to install this as like a workflow is that it doesn't go read our calendar of upcoming posts and launches. And so, as it was going, I was like, oh, you always forget this. This is the message I'm sending of like, actually look at everything that's scheduled cuz I have to account but I didn't go to market plan. Um and then it makes a uh it makes a plan as a proof doc. I went and looked at it and I was like, again, I I maybe have 5 minutes in between meetings and I'm like, this is really good. Like, you kind of have every you have the architecture enough that um I want you to like factor in one other change and then just ship the plan to Notion. And the plan it shipped to Notion, I was reading it and I was like, this is basically 80 to 90% of the way there and that's that's not because it I'm relying on the model to come up with our go-to-market strategy, it's that I'm relying on the model to um look at all of the things that we've already said thought about the go-to-market strategy, piece it together and then review it, right? Come with what will work what what's not. There's a lot of important context loading that happens here where like, it knows what our target ICP is, it knows what our goals are, it knows how we think about narrative positioning. And before this was possible, the only thing I could have done was either block off a whole day to sit and do this or get done with my work for the day at like 6:00 or 7:00 and then stay up all night writing this. And this has been such a game-changer for me and the the other part of it that I think I found is really helpful is that I I don't make this plan for humans. I make this for humans and agents and primarily for humans to understand through agents. And so when I send it to the team working on the go-to-market, they can read it and it's like digestible to humans, but the the thing that it's really helpful for is like it's the full plan sectioned off all in one. And so Brandon, our COO, who's like deep in this product, can ask his plus one, can ask Coda, you know, it's called code, like "Let me know what Austen's plan is. Like summarize it for me. Let me know the business case." Brandon has to come up with the pricing modeling for the plan, so he can work with an agent against the plan. And um as someone who's spent so much time in my career thinking about like literally how the the proposal or go-to-market document looks like, how is it going to look when I present to the CEO, like this two-page plan for for like a for a budget I'm asking for, like is it going to make sense to their eyes and like really fine-tuning stuff, giving up on that and just saying like, "Is the plan really good? And is it going to make sense to like Dan's agents if he approves it?" Um for to me, it makes me work faster, it makes the work better, it means that I don't have to think about all this like kind of dumb stuff that doesn't matter. Um that like it's to me a much more like powerful and fun way to work. \>\> I totally totally agree with that. I you said so many things that are interesting there. The first one is just normalize sending agent documents around. Um and that's why we have proof. Uh it's just such an easy way to send the markdown documents that we generate to each other and and to review them together. And it's like there I think there's this whole strand of AI stuff that's like make AI write in your voice. We even do this with spiral, but there's this other strand of just like normalize AI writing cuz I would actually prefer to read your agent's writing than your writing in a lot of cases because I know that it's just easier for you to get all that that thinking together in a format I can read if you if you have your agent write it. The the I care about is do you stand by it? Have you thought about it? And if I talk to you about it, will it be clear that if I talk about a particular bullet point in it like you've thought that through? And as long as we have the trust that that's going to be the case, then I absolutely prefer the the agent version. \>\> In the future, humans face a new problem. What do you do when your computer is doing your work for you? One answer, think a cloud walk. An idea by Avery. \>\> \[music\] \>\> Avery, the only subscription you need to stay at the edge of AI. \>\> Totally. Like my friend Rachel Karten, who runs the great like Substack newsletter Liquid Bio about social media, had had a really good piece this week about frustrations for people working in social for like every like this pressure they feel that everything has to run through AI and the quality going down. And one reason why is that there's that dichotomy of like what do you actually stand behind? Like are you running something through AI and you like you maybe your manager did it and they don't even know what it what it said. And the thing I love about working at Avery is like you you show up to a meeting, you you've like shared an AI written document ahead of time, and the expectation is that you're going to stand behind all of it. That someone will ask you a question of what's in that document and you if you say like, oh, I didn't even know that was in there, it's like you're you're you're exposed, right? But the other nice thing is that we continue to keep investing in skills and workflows and tools to kind of ensure that never happens. Like I have rules inside of this project file to be like if don't don't add anything that I haven't like said in another context. I want your suggestions. Send your suggestions to me in the chat, but don't put it in a in a document. and like depending on how big the context gets, these models can follow or not follow those rules, which is another reason why I always leave Codex for that final review before it goes to the like humans I work with. \>\> Yeah. And I I think that that that last thing that that that I want to point out that you said is like a lot of the time that you spend working is about taking thinking you've already done and putting it into a form that other people can read and consume. And the important part is doing the thinking. There There is something obviously about like I love writing. Writing is a good way of thinking. Um and sometimes you actually want to do the writing yourself because you want to think about it for certain types of things and certain types of people, but there's a lot of stuff like company strategy where a lot of the thinking happens out loud in meetings. And there's also times like for example, I'm writing something that's sort of like a it's like a retrospective on the last three and a half years of AI and like where I think we're going. And that's so hard to sit down and write, but it's much easier to just like dictate. So I just took a monologue note where I was just like saying stuff and it I'm using the AI to help me like figure out what I'm really trying to say. And And in those cases, I think it's just so nice to record stuff, give Codex access to everything, and then just have it spit out a strategy doc and go through it to make sure it it's stuff you agree with, but it's um such a time saver and especially if you're someone who like us and or like me, like you're in meetings a lot and so you don't necessarily have huge chunks of time in your day to like go do a big strategy document cuz you're just trying to stay on top of whatever's happening. It helps you do that in the cracks of your day and do a lot of that thinking and I just I love it for that. \>\> Yeah, me too. Um I want to show one more thing before we get into more questions cuz this is like I wanted to show kind of like a a more like mix of knowledge work and engineering-y stuff that like would never have been possible without these kinds of tools and that I really love Codex for, which is I've been rebuilding our um KPI tracker every week. Um I'll just like show it here for a bit. So um we have so many different parts of our business at uh at Every. And uh it's very difficult to get all of those um data points in one source of truth in a traditional tool, like even PostHog, which I really like and a lot of our data runs through it, to get one dashboard that is again both human and human and agent-facing that is up-to-date with all of the metrics we care about. I I haven't found a great solution for just like, you know, going to PostHog and having it having it do it. So um we I've been rebuilding our KPI sheet inside of Notion with the goal in mind of any anyone can point their agent to look at it and see how our new paid subscription trials doing, how our page views doing, how is uh Monologue iOS MRR doing, all versus plan, all of this stuff. Because one, it helps you work as a human, but it also really helps you automate agentic work so that you can say like, if your agent sees that we're tracking behind on SEO for a keyword we should be winning on, they can go just like ship a bunch of landing pages for us to try to win more on it if the if the source of truth are good. And so I have been doing this big kind of like to me complex uh workflow problem in Codex of let's build the sheet together, let's have it live in a Notion database that all of our agents can point at, and I've done a bunch of different versions of it. The first version was like, can Codex one-shot this, right? Like it has all the API keys, it has everything. I'm happy to give it the context on like how we measure MRR and everything. And each time it was like a little off. It was like maybe 5 to 10% off of the formatting, the numbers, the framing. And our MRR our our MRR number can't be 5% off. Like we can't run a business where the source of truth is even 3% off. It has to be just exactly right. And so the thing that I forced myself to do and it's weird. Now I'm like it feels so stupid that I have to do this, but it makes sense is like I'm going column by column end to end to ensure each column is exactly right and defensible because it's the only way that we can run and grow the business reliably. And especially the only way we can we can confidently unleash agents to go take actions against what's happening in that KPI sheet. And um it's it's like it's so interesting to me that I'm frustrated that I have to do this that the that the model can't do it for me. Um but it's just because of how like powerful these models have gotten that I expect it to be able to do it. But um and this is a thing where I'm like, you know, it's using um Notion's Workers Tool, which is this like dev tool to build always-on tool calls of our Stripe, of our social. Um it's like creating little scripts and stuff. All stuff I don't really understand, but I understand the outputs. I understand that the output is a Notion database that updates every 6 hours with all of our metrics. And it's just nice that I can do that and I don't need to hire a consultant to do it or like I don't need to like um uh yeah, take away from our uh like our engineers' times that that work on our data. Like I can do this now. And I can do it just by like prompting the model and understanding how the metrics are supposed to work. \>\> It's amazing. Uh I can't wait. Is the Do you think it'll be ready on Monday? \>\> It'll be ready on Monday. yeah. \>\> Feeling really good. \>\> Cuz we've been I mean just having It turns out that figuring out how much money you're making and how much you've grown is truly a philosophical question, you know? Um and you actually do need to like go in and like set that frame. Um and so we've been dealing with an outdated sheet cuz it's like it's pulling numbers, but is are the numbers correct? You know, even even outside of AI. Um and and there's no one way to for example measure MRR. You just want to do it the same way every time. So you have to decide. And uh that's kind of it's kind of wild that it's like almost impossible to tell how much money you've made in an objective way. You have to just like pick. But anyway, uh that's just the way my brain works. I want to say before we get into questions, one other thing that I used this for that would it like blew my mind from a knowledge work perspective is recruiting. So, we're hiring a lot. And we were looking for an L&D head of L&D, someone to help us run courses. And there's this company in New York called General Assembly, and when I think about people who've run like really great courses about technology to teach people how to get hands-on with like programming or design or anything like that. Like they're the company that I think of from the like 2010s in New York. And so I my theory was that if we're hiring for someone to do like build our courses, they would probably have a good person probably would have worked at GA. And Jason, yes. Um and I think GA's quality has gone up and down, but at the beginning they were amazing. Um and uh so what I did was I just said to Codex, "Hey, like can you find Can you just get a list of GA alums? I'm like hiring an an L&D director. And then I want you to filter and sort the list by people who have subsequently gotten into AI. And it did it. Like it just gave me a list of people. The first one I clicked on, it was like I was like this guy is perfect. And then I looked and he followed me on Twitter. So I just DM'd him. And like I don't know if we're going to end up working with him, but like it was just one of those holy light bulb moments where normally what we're doing is sorting through a ton of applications and like trying to find the right person. I We're still going to do that, but especially for any kind of like outbound effort, it can kind of find that needle in the haystack that you're looking for really, really well. So I highly, highly recommend. Um Okay, we've got about 10 minutes left. Um and I want to take some more time for questions. So if you got a question, please uh uh please raise your hand. One thing that we have not gotten to actually is that if you're here, you are getting Codex credits. Uh Austin, do you want to uh go through that really quick? \>\> Yes. So uh OpenAI has given us a code I'm about to drop into the chat for 250 attendees of this camp to get a free month of ChatGPT ChatGPT Pro Lights. That's about a $100 value. Um and you can redeem it at this link that uh we will drop in the chat right now. \>\> Sick. So \>\> Uh Dan Dan, I'm actually going to I'm going to Slack it to you so you can drop it in the chat because for some reason I don't have access. \>\> Okay, I'll do that. Um so yes, this is this is our gift to you as every subscriber. We try to do stuff like this all the time. We've been We've given out I think we've done um cursor credits. We've done We've done a lot of a lot of other stuff. We have more stuff like this coming. So we just want you to be able to try these tools, be at the edge with us, and um we just love having you as a subscriber. So here is the link. It's $100. Oh, Notion, yes, we did give out Notion. It's $100. Um and uh check it out. We will send it out in an email. Maybe we actually we may not send it out in an email because it's only 200 limited to 250 people and that's pretty much exactly the number of people who are here. So, if there's any left, we will send it. Um if there is one person there's 251 people here, so there's uh So, if uh that person if you let us know, we'll figure something out for you. Um interesting, not available on my plan. Um Okay. Uh we will have to deal with this. Uh let us let us figure out what to do what to do here. So, correction, this is only if you do not have a plan. This is for new users and we'll try to we'll try to get something for existing users and send it out as soon as we can. Cool. All right, let's do some questions. Um Rich, please ask your question. \>\> So, I I saw at the beginning you were using Compound Engineering as kind of part of your workflow. Are you using kind of like the off-the-shelf plugin or is there tweaks to it and kind of where does that work and maybe not work when you're outside of the you know, kind of code creation workflow? \>\> So, I I find that there's no um overwhelming need to fork your own version of Compound Engineering. I used it for a long time. Um for all of my knowledge work, and it was extremely powerful for me. And then, um maybe about 2 months ago, the the main thing I noticed was reading the agent's response to especially the review stage of watching the reviewers that Kieran and Trevin had built that are very specific to engineering. It it it was like, oh, this like the the thing you'll see the agent do is say like, um I'm supposed to go through through this review step. It looks like it's designed for engineering. It's thinking about security and front-end design when this is a go-to-market plan. The agent will then like change the path. The agent will be like, I'm going to review this for something else rather than reviewing it for security. And so, the thing that I did was I went and forked a version of it that is actually, um publicly available on our GitHub called compound knowledge, which is, uh, built exclusively for me taking the compound engineering plugin, which is also public and you can go fork, and going inside of I think I started in Claude code, now I updated it updated it in Codex, and saying like, I want to tweak this to general knowledge work. And this is the thing I was I was referencing around the like reviewers being much more specific to knowledge work around like strategic alignment and and data accuracy. I think more than anything, this is like a really fun way to learn, um and a what fun way to like kind of like push yourself on using models. You're welcome just to go use this one. We'll include it in the in the follow-up email to the camp. Um but I think it's a cool like I learned a ton just by doing this. I had never made like a plugin like this before. Um and to make your own version of say you do like social media marketing and you want to make sure all the reviews go through your style guide, your like past performance, um I I got a ton out of operating this way. If you just want the compound engineering to make your work better, it absolutely works really, really well for knowledge work just kind of out of the box. \>\> Got it. Yeah, no, interesting particularly using kind of all the the inner step pieces like compound that that's still a apparently a valuable step for you. \>\> Yeah, other compound step is is really valuable. We have um, inside of our notion a go-to database of after you're done with a session, um, you can send the learnings from the session to actually a team-wide shared compound source of truth. Um, whenever I'm done with any session in Codex or Cloud Code, the agents are instructed to ask me, should we compound this, save it somewhere for the learning, and should we turn any workflow from the session into a skill so that we um, can just do it automatically each time. \>\> Got it. Cool. No, I'll check that out. Thanks. \>\> Cool. \>\> All right, Rory, please introduce yourself and ask your question. \>\> Hi, my name is Rory and I'm in your head. Um, is it Are there anything uh, about the way you work at every um, like maybe taking some time after meetings, like ending them a few minutes early so that you can do those things that um, you'd recommend to teams that are adopting workflows like yours? Is that clear? \>\> Uh, yeah, I I think so. Like, um, but to say it back to you, like what what I'm hearing, which is like a very real challenge here, is that um, it's the it's so exciting and tempting and alluring to like spend a lot of your day playing with stuff. Also spend a lot of your day continuing to push on like if I just get this automation right or this tool right, my work is going to be at like 100 times better and easier. And I actually find myself a lot like on a lot of days spending most of my time not in meetings trying to build really good tools and automations that work well and not making the time to do the actual like tasks that have to push the the business forward like like uh shipping the social posts for the day or or whatever. And I I don't really have like an awesome answer for it that outside of the fact that like the the playing around one is like kind of core to how we operate at every. It's it's a thing that Dan like pushes all of us to do. It's one reason why I love working here. It's also like to me the best way to learn and and makes me better at everything I do. And then um the the the only kind of like guidance I've given myself is that like these automations in Coda X keep me on track to get the work done so that when I'm too deep in um in like playing around and building this like there's like a social automation tool I'm working on that I've been like deep in for a while. Um the Coda X automations make it so that I like, you know, make sure Brandon gets what he needs for this like um some like biz dev plan we're we're doing. I I do find myself over indexing on learning and and playing because of how exciting and powerful the models have been and that more after like continue to pull myself into the the like required day-to-day tasks and the the urgent stuff that's happening. \>\> Yeah. And I I also sort of read your question, Rory, and you tell me if this is wrong, but as like how do we do more of the AI stuff, the more of the playing even even get started on this stuff in our day-to-day uh if we're like busy all the time. And I and what are the organizational practices that we have for that? And yeah, I just think like Austin said like it's just like a culture it's a cultural thing. Um we just love playing around and that's like that's part of our job. Um and I think there's this there's this thing happening right now where the tools and the workflows are changing so fast that just focusing on how your job currently works you you can run as fast as possible and someone using a new tool with a new paradigm and a new workflow is just going to beat you by default. And so if you just give yourself some time to play around it may feel like a waste of time, but you're leveling yourself up to a different game at a different level. And I think that's really important and some of the organizational practices that we have to help people do that are really around so one of the things we do twice a year is called Think Week. Um and we just literally don't do any of our day-to-day work and we just spend a week together just like playing around with new stuff and building stuff and learning and and being together and you don't necessarily have to do a whole week of that, but um I think it's really good to maybe do that once a quarter for a day or something like that. Um and just give people the the time and space to if if you can. Sweet. Um all right, y'all. So that is our program for today. Thank you for coming. Uh we love seeing you. We love doing this with you. Remember Every is the only subscription you need to stay at the edge of AI. We would love it if today you would go tell one of your friends to go subscribe to Every. Um we want to get more people in here. We just think we're we're right at this amazing point in history where we get to surf ride this big wave together and figure it out together together. And um please please tell your friends. See See \>\> Thanks, y'all. \>\> Oh my gosh, folks. You absolutely, positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure, unadulterated knowledge bombs about ChatGPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show, it's a journey into the future with Dan Shipper as the captain of the spaceship. So, do yourself a favor. Hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely, hopelessly in love with you.