Modern Creator
Greg Isenberg · YouTube

Inside Codex Sites: Full Tutorial Before Public Launch

A 25-minute live build that reveals why Codex Sites is an agent harness, not a Replit clone.

Posted
today
Duration
Format
Tutorial
educational
Views
783
58 likes
Big Idea

The argument in one line.

Codex Sites is not a no-code builder -- it is an agent harness for apps that update themselves, and the three concepts that unlock that are memory, safe actions, and skills.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You already use Codex daily and want to extend it into a deployable product, not just a chat session.
  • You are building internal tools or solo apps and want Codex to keep operating them after launch without manual editing.
  • You understand the basics of databases and auth but do not want to leave the Codex ecosystem to wire them in.
  • You have tried Replit or Lovable and want more control over the agent update loop.
SKIP IF…
  • You want one-prompt, fully managed deployment with built-in auth, payments, and custom domains -- Replit or Lovable is the better fit today.
  • You are brand new to AI coding tools and want the lowest-friction first experience.
TL;DR

The full version, fast.

Codex Sites trades ease of first deploy for tighter integration with the Codex agent loop, meaning apps can update themselves from a new chat with no manual editing. The tutorial builds a Startup Ideas OS across six prompts -- shell, Cloudflare D1 storage, safe actions (named mutations Codex is allowed to call), a skill so future chats know how to operate the app, a save-gate checkpoint, then a loop-prove that opens a fresh chat and autonomously adds a card to the live board. The host's actual thesis: most people will use Codex Sites for personal pages, but the real unlock is making products Codex can keep operating for you.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →
Chapters

Where the time goes.

00:0001:17

01 · Intro and Episode Agenda

Skeptical hook (is this just a worse Replit?), then six-prompt promise for the tutorial.

01:1704:33

02 · Codex Sites vs Replit and Lovable

Replit/Lovable = full-stack one-prompt. Codex Sites = agent-native, missing auth/payments/vault but deeper Codex integration. Internal-only, no custom domains yet.

04:3305:08

03 · The Build Plan: Startup Ideas OS

Kanban board: inbox, researching, validating, building, killed. Cards with idea, buyer, pain proof, next step, score. Six prompts planned.

05:0807:02

04 · Prompt 1: Build the Shell

Invoke Sites as plugin, use realistic sample data, save for review (do not deploy). Key tip introduced.

07:0208:54

05 · Plugins Worth Using

Plugin directory tour: Figma, Canva, HeyGen, Remotion, GameStudio. Insight: mini-games as attention-funnel into a core product.

08:5409:21

06 · First Board Review

Live Startup Ideas OS board shown -- clean, minimalistic, good enough for v1.

09:2110:56

07 · Prompt 2: Add Memory and Data Model

Ask for data model before code. Codex picks Cloudflare D1. Safe actions concept introduced.

10:5613:25

08 · Prompt 3: Create Safe Actions

Named mutations only (add, update, move). Non-technical tip: ask Codex what safe actions your app needs.

13:2514:51

09 · Prompt 4: Create the Skill

Skill = reusable instruction manual. Tells future chats how to operate the app. Validation passes.

14:5116:29

10 · Prompt 5: Save-Gate / Checkpoint

Video-game checkpoint analogy. Save as v1 review, do not deploy. Build passes, D1 confirmed.

16:2918:10

11 · Prompt 6: Prove the Loop

From a NEW chat: skill loads, safe action called, idea added to inbox. No raw SQL, no manual deploy.

18:1020:28

12 · Publish, Auth, and Live Updates

Deployed to Codex URL. Auth built in. Prediction: custom domains coming soon.

20:2822:40

13 · TLDR: Memory, Safe Actions, Skills

On-screen summary slide. Memory = app saves data; Safe Actions = Codex uses approved mutations; Skill = instruction manual.

22:4024:41

14 · The Real Unlock and Closing

Wow moment: open new chat, add idea, live site updates. Real unlock = products Codex keeps operating autonomously.

Atomic Insights

Lines worth screenshotting.

  • Codex Sites is not a Replit competitor -- its moat is autonomous post-launch updates, not ease of first deploy.
  • A website without persistent storage is a demo. Codex Sites ships without memory by default -- you have to ask for it.
  • Safe actions are named mutations Codex is allowed to call -- they prevent the agent from writing arbitrary SQL to your production database.
  • Skills are reusable instruction manuals that tell future Codex chats how to operate your app without reading the codebase.
  • Always say 'save for review, do not deploy' in Codex Sites -- otherwise it deploys before you have reviewed anything.
  • Save-gating is the Codex Sites equivalent of a video game checkpoint -- commit a named version before going live.
  • The real unlock is opening a new chat and saying 'add X to my app' and watching the live site update automatically.
  • Most people will use Codex Sites for personal pages; the edge is using it to make products agents keep operating for you.
  • Non-technical founders can skip guessing what safe actions to create -- ask Codex what they should be based on the data model.
  • GameStudio plugin enables attention funnels: build a mini-game tied to your content topic, funnel players into the core product.
  • Cloudflare D1 is Codex Sites default durable store -- the agent picks it automatically when you ask for persistent storage.
  • The loop-prove is the acceptance test: if a fresh chat can operate the live app via skill and safe actions, the build is done.
Takeaway

Three layers that turn a Codex build into an autonomous product

WHAT TO LEARN

Memory, safe actions, and skills are not optional polish -- they are the difference between a demo that works once and an app Codex can keep operating without you.

02Codex Sites vs Replit and Lovable
  • Replit and Lovable win on ease of first deploy; Codex Sites wins when you want the agent to keep operating the app after launch -- they are solving different problems.
  • Custom domains and a vault for secrets are missing in Codex Sites today; if you need them now, Replit or Lovable is the pragmatic choice.
04Prompt 1: Build the Shell
  • Always tell Codex Sites to save for review and do not deploy -- without that instruction it deploys mid-session before you have validated anything.
  • Asking for realistic sample data in the first prompt saves an extra round-trip; empty boards do not reveal layout problems.
07Prompt 2: Add Memory and Data Model
  • An app without persistent storage is a demo regardless of how good it looks -- memory is the difference between a prototype and a product.
  • Ask Codex to show you the data model before writing any code; seeing the schema first lets you catch structural mistakes without rolling back.
08Prompt 3: Create Safe Actions
  • Safe actions are a permission boundary -- they limit what the agent can do to named mutations, so it cannot write arbitrary SQL to your live database.
  • Non-technical builders can skip guessing what safe actions to create -- prompt Codex to list them based on the data model it already showed you.
09Prompt 4: Create the Skill
  • A Codex skill is an instruction manual for your app: without one, a new chat has no idea how the board works or what commands are valid.
  • Create the skill before save-gating -- it is part of the working product, not documentation you add later.
11Prompt 6: Prove the Loop
  • The loop-prove is the acceptance test: open a fresh chat, send one command, and confirm the live site updates without any manual editing.
  • If the loop fails, the skill or safe actions are incomplete -- fix them before calling the build done.
14The Real Unlock and Closing
  • The wow moment the host describes -- typing add this idea to my Startup Ideas OS in a new chat and watching the live site update -- is only possible if memory, safe actions, and skill are all wired together.
  • Most people will use Codex Sites for simple pages; the builders who wire up all three layers are the ones building products, not demos.
Glossary

Terms worth knowing.

Codex Sites
OpenAI's Codex feature for building and deploying web apps from within Codex, with native agent-update capability. Apps built with it can be operated and updated by future Codex chats without manual editing.
Safe actions
A predefined set of named mutations (add, update, move, archive) that the Codex agent is explicitly allowed to call on an app. Prevents arbitrary database writes and constrains the agent to an approved API surface.
Skill (Codex)
A reusable instruction file stored in Codex that tells future chats how to operate a specific app -- board structure, valid commands, example prompts. Functions as an instruction manual the agent reads before acting.
Save-gate / checkpoint
A versioned save of the current Codex Sites build before deploying, preserving a known-good state to roll back to. Analogous to a save point in a video game.
Cloudflare D1
A serverless SQL database on Cloudflare's edge network. Codex Sites uses it as the default durable store when the user asks for persistent storage.
Autonomous loop
The practice of opening a fresh Codex chat and verifying it can operate the live app via skills and safe actions, without the builder manually editing anything. Used as the acceptance test for a finished Codex Sites build.
Vibe coding
Informal term for building software primarily through natural-language prompts to AI, with minimal manual code editing.
Resources

Things they pointed at.

01:41productReplit
01:41productLovable
12:42productBolt
10:13productCloudflare D1
21:01productConvex
07:28productGameStudio (Codex plugin)
07:35productFAL.ai
07:36productHugging Face
Quotables

Lines you could clip.

00:00
I saw that Codex launched Sites. And when I first saw the news, I was like, is this just a worse version of Replit or Lovable that's just inside Codex?
perfect skeptic hook -- relatable reaction shared by most viewersTikTok hook↗ Tweet quote
02:33
What's the coolest part about Codex Sites is it updates your app autonomously.
single-sentence thesis, no setup neededIG reel cold open↗ Tweet quote
21:57
Most people will use Codex Sites to make pages. The move is making small products Codex can keep operating for you.
tight TLDR contrast, quotable as-isnewsletter pull-quote↗ Tweet quote
23:31
You could open a new chat and say 'Add this idea to my Startup Ideas OS.' Then show the live site updating. That is the wow moment for me.
concrete wow moment with a specific example -- great demo clipTikTok hook↗ Tweet quote
The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

metaphoranalogy
00:00I saw that Codex launched Sites. And when I first saw the news, I was like, is this just a worse version of Replit or Lovable that's just inside Codex? But the more I dug into it, the more I realized it's actually worth understanding how to use it and how to get the most out of it.
00:18So today is a full tutorial on basically what the big announcement is around Codex Sites, how does it compare to the competition.
00:28We're actually gonna go and live build using codec sites. And through that, I'm actually gonna teach you, uh, basically the best practices for using codec sites. So by the end of this episode, if you stick around, uh, I'm gonna teach you how to build a how to build a shell using Codex Sites, how to add memory using Codex Sites, how to add what's called safe actions using Codex Sites, and I'll explain what that means in in the episode.
00:55How to create skills with codec sites, how to save gate using codec sites, and how to prove the loop using codec sites with which makes it, uh, work autonomously.
01:07Let's get into it.
01:17So what is the difference between Codec sites and Replit and Lovable and and stuff like that?
01:24So I will say, Replit and Lovable and and and those and those products are really good if you wanna like one prompt something and it has an editor, it has a database, it has server, it has hosting. A lot of them have connections to domains as well too.
01:41Like, can just register a domain within it.
01:45Um, so that's really amazing. Um, but Codex Sites, if you live in Codex, and I'm starting to live in Codex.
01:54I mean, I think it was about a month ago I had Riley Brown on the podcast, and I I basically said, like, hey, convince me to use Codex. I'm in the Anthropic ecosystem. Convince me to use Codex.
02:04And now I will say it's a part of my daily driver. Yes. I still live in the Anthropic ecosystem, but I'm also using Codex.
02:13And, um, I think that if you are in Codex and you are putting just all your context there, what's really cool about Codex Sites is, uh, you can actually use that to go and build your ideas, your apps, and stuff like that.
02:30And what's the coolest part about Codec Sites is it updates your, uh, your app autonomously. So what does that mean?
02:38It means that, uh, you can have it, uh, basically create a personal website, say, and, uh, you know, and let's say, for example, go to my personal website.
02:51Um, this is a static this is a static website. So I actually have to go in and and basically, you know, a 100 a 158,000 people enjoy reading my newsletter, but, like, when it becomes a 160,000, I have to go and update it.
03:08But with Codec Sites, you can actually have it so it automatically updates these things. It automatically creates, for example, guides based on my content.
03:17It automatically, you know, adds to different companies I work on. So it basically is this whole idea around autonomous product building.
03:25I've talked about this on the podcast before, but the idea around you're going to have, uh, in the future, and and the future is now the present, basically agents going and updating products autonomously.
03:38But the thing is, it is missing auth. It's missing databases. It's missing payments.
03:43It's missing email sending. It's missing analytics, and it's missing a vault for secrets.
03:48So a lot you know, if you want something more simple, everything is in there. You're gonna wanna use something like Rapid or Lovable. But if you understand a little bit about auth, if you understand a little bit about database, payments, stuff like that, and you live in the Codex ecosystem, and you are really interested in this whole idea around autonomous apps, Codex sites are really, really, really cool.
04:13And I also say one thing. Today, you can't publish these websites, at least not to my knowledge, you can't publish these websites on your own domain and go and deploy it publicly.
04:25These are internal apps that you can share with your team right now. But obviously, that's gonna change very, very, very, very, very soon. So let's get right into it.
04:35So I I was thinking that we can create a startup ideas OS. So basically, live board with columns, inbox, researching, validating, building, and killed startup ideas.
04:47And each card can have an idea, buyer, pain proof, next step, and score.
04:52So this is something that, you know, could be an internal internal tool that I would use. And we're gonna basically try to build it in six prompts. And like I said, as I build it, I'm gonna share, uh, sort of the big takeaway.
05:07So the first, uh, first thing we're gonna do, let's go open Codecs.
05:13So I, uh, created a new project here. I opened a, uh, a new chat and, you know, the first thing is, you know, if you wanna actually use sites, you have to invoke it at sites.
05:27So it works basically as a plugin. So I said build a startup OS. Um, I'm gonna go ahead and send it.
05:36Um, and, yeah, it works as a plug in. You invoke it there.
05:42And like I said, not everyone has access to it, but I would imagine that access is probably coming really soon for for plus plus people.
05:53And by the way, I have no affiliation with OpenAI. I just think this is a really interesting product, and I wouldn't be making this episode if I didn't think that you can get a lot of value from it too. So, um, what is the insight I have for you about this particular prompt is that, you know, um, well, there's a few things.
06:14Uh, Yeah. Invoke. The way to to to get it going is you wanna invoke sites.
06:21You wanna actually ask it to use realistic sample data. Um, you also wanna save it for review. So I noticed that, um, sometimes when I I prompt it an idea, it'll just try to deploy it.
06:35So a really good tip is to save for review, do not deploy. Um, and this basically unlocks building a real product service, not a homepage.
06:44So, basically, I what I'm trying to do by building this is to build that whole idea around autonomously being able to edit it and and work on it, um, because that's sort of sort of the dream state for me. Um, so let's go ahead and go back to Codex.
07:03Okay. While that is being worked on, I just wanted to say that if you if you go to the plugin section over here, so you click plugins, there's a bunch of plugins that are going to help you make your sites a lot better.
07:21Right? So, um, you know, a lot of people aren't using these plugins, um, but think about, like, Figma.
07:28Think about Canva, think about HeyGen for avatar videos, think about Remotion.
07:35These are a bunch of plug ins that could make your ideas for your sites a lot more interactive, a lot more a lot more interesting. Uh, another really interesting one that's underrated, I think, is GameStudio.
07:50So, um, you know, one of the biggest problems that people have with vibe coding and agentic engineering is they build something and no one no one's no one's going to it.
08:03Well, you know, if you build games, for example, that actually generate buzz around what your the bigger you know, the the product you're building, that's a way to get attention, and then you bring those people into your core your core product.
08:20Now, of course, that needs to make sense for whatever it is you're building, and this works better with consumer apps. Right? So you go and create a, you know, a paperboy competitor that's fun to play, that you can kinda, like, news jack based on a particular news topic, and then you bring those people into, you know, whatever your core experience is.
08:42Maybe it's an email newsletter or who knows. But the point is, like, I think a lot of people are sleeping on this idea that you can go and create little apps that could bring them into the core experience. Alright.
08:55So it's built the board. I think it looks pretty pretty clean.
09:02Got the inbox over here, researching, validated, building, kill.
09:07Definitely not the most beautiful, but clean, minimalistic, good enough for a first version.
09:16So let's go back and keep going.
09:21So how do we add memory? Right? Because on codec sites, it's it's not gonna add memory without us basically getting it to to prompt it.
09:31So let's go ahead and go back into codecs. I'm copying this prompt. So I say, add persistent storage so ideas save between visits stay saved between visits.
09:44Before coding, show me the data model in which records action the app needs. And this is by the way, these prompts are similar prompts that you can use for when you're using Codec sites. Um, so it says, I'll use the site's, uh, building guidance here because this is a hosted app with storage.
10:01Since you asked for the model first, I'm gonna inspect the existing schema and then layout the proposed records and mutations before touching code.
10:11So it says it's actually gonna use Cloudflare. Right? I'll use Cloudflare d one as the durable store with one main record type ideas.
10:18This already matches the project's current direction, and OpenAI has d one DB data model. So then it goes, and if you if you see here, it says the record the apps need.
10:30So it says it needs an idea, one one card on the board, it's thought it stores the full card content, the current column, it store, uh, needs the owner email, and then these are the actions that the app needs.
10:42List ideas, add idea, update idea, move idea, scoring the idea, archive the idea, ensure ensure seated. Um, and then it says, I'll wait for your go ahead before coding the persistent path pass.
10:57So this is the the whole idea of safe actions, which are, you know well, before I get into that, why add memory?
11:08Well, because we need for this to be a productive piece of software, we needed to remember things. Right?
11:15So we wanna make sure that we've got that in there. But the whole idea around safe actions is is gonna be it's it's an unlock because we can be in other chats. And because we live in Codex, we can basically say, hey, you know, maybe I'm, like, ping ponging with Codex, and then I it hits me back with a an idea, startup idea, and I'm like, oh, this is really good.
11:36Let me just let me just add add so let me just at add idea and it it'll directly add it to the application, which is so cool. So, uh, you're gonna wanna create safe actions because, you know, in order to get the most in in order to get the most out of it.
11:53So I'm just gonna go ahead and and copy that prompt, create safe actions for add idea, update idea. And and the reason, you know, it's important to that I, you know, I I I said show me the data model and which record and actions the app needs is, you might not know what safe actions to create, so you you might as well act ask Codex, and it'll give it to you right there.
12:14So for people, nontechnical people especially like that, you might not know what your safe actions are, you just ask Codex what what they should be. So it says, I'll use the site's building workflow again since this is a worker d one service. Um, I'm gonna inspect the current action routing and tighten it into an explicit safe action boundary so the agent can only call named mutations rather than the arbitrary SQL.
12:39So you're starting to see that codex, it does feel a little more technical than your your Lovable or your Replit or your Bolt. But if you can stay with it, the output, I think, is super, super valuable.
12:55Like, the idea that you can have it self self update, the the idea that you're in Codex and you can bring in all this stuff from Codex, the idea that you can go into the plug ins and and just use all this stuff from the get go.
13:10Like, these are some of the I mean, FAL, AI and image generation, like, so much stuff.
13:18Hugging Face, these are all open source models, like, it's all built in from the get go. It's pretty sweet. So that's completed.
13:26So the next thing we're gonna wanna do is create a skill. So we're gonna copy this here so that the chats, our future chats are gonna know how to use this app. So create a codec skill called startup ideas admin.
13:43It should explain how to read the board, how to add ideas, how to move cards, how to score ideas, and include five example commands. This is something that I think that is you know, I I started seeing people use codec sites, and they're not using Skills.
13:59And I'm like, this is one of the greatest parts of of Codex Sites is the ability to create these skills. So you might as well might as well, uh, start adding some skills to get the most out of your application.
14:13So it says, I'll use the skill creator guidance for this since you're asking for a new codec skill. I'm gonna read the skill instructions and create StartUp ID as admin in your local skill directories with the board action workflow and example commands. The skill name is already valid, and the scope is clear.
14:29Operational guidance for this board, and its safe actions. So it's gone ahead and created the codec skill.
14:38It's done everything we've asked for, reading the board, adding ideas, updating cards, moving cards, scoring ideas, archiving ideas, five example commands, and it says validation pass.
14:49The skill is valid. It's going to be helpful for you to actually save gates.
14:56What do I mean by that? I mean, you know, the way I think about it is like a video game. Right?
15:00Like, you have, like, checkpoints as you go through a video game, and sometimes it'll auto save, and you're you're happy it auto save. But the thing with Codex is it doesn't auto save.
15:10Um, so it's helpful to go in there and just say, hey, just do a checkpoint here before a live URL. So, uh, let's go ahead and go back into Codex, and I'll say, save this as v one review.
15:24Do not deploy. Confirm build status, storage choice, access setting, and the exact version I should review. And I'll include, by the way, all these prompts, I'll include in this show notes in the description for for people to download, um, so that they can go ahead and and and as they're building codec sites, they could they can go and use some of this stuff.
15:45So it says, I'll use the site's hosting workflow for the save only review session. I'm gonna validate the current build, confirm the storage access config, commit the latest safe action changes, and then save a site's version without deploying.
15:58There you go. So prompt is done. It's saved for review.
16:01It isn't deployed. The build status is passed. Um, the storage choice is still d one.
16:06The access setting is custom.
16:10One thing to note here is at the top right, you can see, like, your whole environment here. So you can see all the different changes you've made. You can see if you're working on locally or you're connected to Codec's web.
16:21Here, I'm connected to the main branch. And so, you know, it's just all all up here on your top right.
16:30The last prompt I wanna do is the whole proving the loop process. So what I'm gonna do is I'm gonna grab this prompt right here, and I'm gonna say, in a new chat, important important that it's in a new chat, use startup ideas admin to add, here's the startup idea, AI agent SEO grader for local businesses, put it in an inbox with a first pass score in the next step.
16:53And let's see if it actually can do that. Does it open up a new chat or do we actually have to go and open up a new chat? Let's see.
17:01It says, I'll treat this as two pieces. Use the new startup ideas, admin skill instructions, and create a fresh chat thread to do the actual board update there. Yes.
17:10I'm loading the skill and finding the thread creation tool now. The skill says to only use oh, look, it's gone and opened.
17:19It's gone and created the chat. I can open it right here. So it says, I'm using the skill.
17:25Um, I'm gonna find out how the local app is meant to run. I'm gonna read the board.
17:29I'm gonna add the inbox card and the safe action and verify by reading again exactly what I wanted to do. The project is clean and looks like a Next style app with a worker and a verifiable script. I'm checking the API implementation now so I can call the same local service the app expects.
17:44The app expo exposes exactly the safe route the skill expects. Yeah.
17:49So this is doing exactly what I want it to be doing, and let's see if it works.
17:56Alright. So the prompt is done. It says done using only the safe board API, which is what we wanted in the action layer.
18:02There's no raw SQL SQL, no generic database rights, and no deploy. It's gone and did it. It went really, really fast.
18:11So now I'm gonna say, can you publish now I'm gonna say, can you publish website and see what happens.
18:20So it's gone ahead and, uh, deployed it. You'll notice that it's this crazy URL, and that's a downside of Codec sites at the moment.
18:30I think this is gonna change. That's my prediction. It's gonna change hopefully soon, whether you're going to be able to have custom domains.
18:37But I can go and open it in Codex browser on the right here. It's got auth built in here. I do have to sign up.
18:44So let me go ahead and sign up. So as you can see, this is exactly what we wanted.
18:51We've got the inbox, the researching, the validated. Um, if I make it bigger here, I can see more the building, the killed.
19:02And it's got everything I've asked for. Right? Um, now, if I want a new idea, I can go and actually create a new idea or again, I can use a safe action to go and do it or I can have it run an automation, like, every week to automatically add to this board.
19:20So we've seen that you can actually do that. Um, I can go more in-depth in a future video if people are interested in, like, how to do automations and cron jobs and stuff like that.
19:29But you can basically just, you know, ask Codec Sitesco and do that for you.
19:36And I think what's really cool is that, like, you know, if you listen to this channel, a lot of us are interested in creating startups that, you know, we don't need big teams to to manage.
19:49Right? And this whole what's so cool about this is we're getting to this with Codec Sites. We're getting to this if you can create agents that go and automatically update based on, you know, different, uh, criterias, then and then and it were and these are products that work and are valuable to people, then, um, that's sort of sort of the dream.
20:10So this, you know, is it the most beautiful website on the planet? No. Does it work?
20:15Yes. Does it look decent? Yes.
20:17Could I get it to a point where it looks beautiful? Um, I can. And and that's that's pretty cool.
20:29So to to to, like, t d TLDR it, like, are the main concepts to understand in within Codec Sites is number one, you're gonna wanna ask it to have memory.
20:44Right? You know, the app saves data, but without this, it's just a demo.
20:48So you're gonna have to ask it to to save memory. You're gonna have to ask it to have a database. A lot of people are starting to use Convex with with Codec sites, so it's something to look into.
21:02I imagine that over time, it's just gonna be easier and easier to use Codec sites, But now you do need to prompt it to do a lot of these things.
21:11The second thing is this concept called safe action. So this whole idea that you can have approved buttons, um, and and through that, you can have it automate, uh, adding, removing, editing your apps, um, so that, you know, you are, as the human being, aren't doing everything, and you don't have to actually edit every, uh, every website or app you create.
21:39You know? I used to run a web design agency.
21:42It was one of my first jobs. And, you know, one of the things I used to say was, a website is a living and breathing entity.
21:51It isn't something that you hit publish and you can just, like, walk away forever. And, you know, we're now in this era, 2026, where the agents are actually doing the the the updating, the editing, the removing, and that's through things like safe actions.
22:10And skills, like use skills with Codex. We saw how how it could be valuable in this in this episode, but a skill is basically this reusable instruction manual so Codex knows how to operate the app later.
22:23So you're gonna need to create skills so you can use the safe action safe actions and so that and obviously, you're gonna wanna store that in the memory so that it, you know, it it it does it safely and and is giving the right information.
22:39So all this to say, the wow moment for me is is building apps where I could do, like you know, in this example, I could open a new chat and say, add this idea to my startup ideas OS.
22:54It shows it shows it in the live site updating. Right? So once it's published, I could just be like, do this, and it's gonna add it.
23:03Or I can have an agent do it, and it's gonna add it. And it's gonna be all on my live website, which is really cool.
23:11Um, I think that a lot of people are gonna use Codex Sites to make, like, personal web pages, to make, like, little apps and stuff like that. But I actually think that the the real unlock here is to make products that Codex can keep operating for you.
23:27That is what's exciting me about Codex Sites is that it it if we can create ideas and apps and websites that are automatically updated, automatically get better, are are autonomous, I just find that so so interesting, and I think this is a trend that's only gonna get bigger, better.
23:44And it was interesting to see that Codec Sites is is leading the charge. So, yeah, this has been a inside look as to how to use Codec Sites, what I think is interesting about it, some best practices.
23:58I hope it's been helpful. I'm just trying to share the most this the new tools that I think are are gonna make, uh, increase.
24:06I'm just trying to share the the new tools that are gonna increase the probability of success for you, whatever it is you're building, and just share the best way to use these tools and give you ideas along the way.
24:18If this gave you an ounce of value, please like, comment, and subscribe. I'll see you in the comment section.
24:25I read every single comment, and I'm I'm rooting for you.
24:30You know, whatever it is you're building, I'm rooting for you, and I can't wait to see what you build. I'll see you on the next one, and thank you for joining in the Startup Ideas podcast.
The Hook

The bait, then the rug-pull.

The knee-jerk read on Codex Sites is obvious: another AI builder in an already crowded field, crammed into a chat interface. But the host came in skeptical and left convinced -- not because it out-features Replit, but because it solves a completely different problem: making apps that Codex keeps operating for you after launch.

Frameworks

Named ideas worth stealing.

05:08list

Six-Prompt Build Sequence

  1. Build the shell
  2. Add memory + show data model
  3. Create safe actions
  4. Create skill
  5. Save-gate checkpoint
  6. Prove the loop from a new chat

A reusable six-step recipe for building any Codex Sites app that Codex can operate autonomously after launch.

Steal forany Codex Sites build, internal tool, or AI-operated product
20:28list

The Main 3 Concepts (host's own summary)

  1. Memory -- the app saves data; without this it is just a demo
  2. Safe Actions -- Codex can only use approved mutations
  3. Skill -- a reusable instruction manual so Codex knows how to operate the app later

The three non-negotiable layers that separate a Codex Sites demo from a product.

Steal forexplaining Codex Sites to non-technical stakeholders or in a tutorial context
CTA Breakdown

How they asked for the click.

VERBAL ASK
24:22subscribe
If this gave you an ounce of value, please like, comment, and subscribe. I read every single comment.

Warm, direct, personal close. Follows the TLDR summary so CTA lands after value is delivered.

MENTIONED ON CAMERA
01:41productReplit
01:41productLovable
12:42productBolt
10:13productCloudflare D1
21:01productConvex
07:35productFAL.ai
07:36productHugging Face
Storyboard

Visual structure at a glance.

hook open
hookhook open00:00
vs comparison
contextvs comparison01:17
prompt 1
valueprompt 105:08
plugins
valueplugins07:02
board review
valueboard review08:54
data model
valuedata model10:56
prove the loop
valueprove the loop16:29
TLDR slide
valueTLDR slide20:28
real unlock
ctareal unlock22:40
Frame Gallery

Visual moments.

Watch next

More from this channel + related breakdowns.

Chat about this