27:01Davie Fogarty Mentors · Tutorial
Learn 90% of Codex in Under 30 Minutes
A live, warts-and-all build of a Facebook-ad tool in Codex, used to walk through seven stages from one lazy prompt to a self-improving AI agent.
September 22ndThe argument in one line.
ChatGPT's Work, Skills, Agents, and Dot combine into a no-code marketing team where reusable skills handle repeatable tasks, specialist agents handle judgment calls, and a persistent assistant coordinates both from anywhere.
Read if. Skip if.
READ IF YOU ARE…
- A solo marketer or small-team lead who wants a repeatable system for content, ads, and landing pages without hiring a full agency.
- A ChatGPT Plus or Pro user trying to understand what Work, Codex, Skills, Agents, and Dot actually do differently now that OpenAI has shipped all five in quick succession.
- A founder who wants an assistant that can run small marketing tasks over Slack or email without being at a desk.
SKIP IF…
- You're not on the ChatGPT Pro plan, since Dot is currently Pro-only and most of this workflow leans on it.
- You need enterprise-grade governance or permissions around AI agents touching company files; this is a solo-creator setup, not an IT-reviewed rollout.
The full version, fast.
ChatGPT's newest pieces, Work, Codex, Skills, Agents, and Dot, combine into a no-code AI marketing team. Grace Leung sets up a branded project folder, teaches ChatGPT reusable skills for repeatable jobs like ad creative and website builds, then creates specialist agents, a growth strategist and a content creator, for tasks that need judgment. Dot, a persistent assistant with its own computer, sits outside the team as an executive assistant reachable through Slack, email, or voice. Messaged with one plain-language request, it routes work to the right agents and skills, then returns a finished campaign: drafts, a posting schedule, a hosted landing page, without the user re-explaining the brand or giving step-by-step instructions.
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 →Where the time goes.
00:00 – 00:27
01 · ChatGPT for your marketing
Grace names the problem: ChatGPT now has Work, Codex, and Dot, and it's confusing which to use. She promises a full AI marketing team build, no coding required.
00:27 – 01:20
02 · ChatGPT Work vs. Codex: where to build
Work and Codex share the same underlying model but target different outcomes: Work for multi-step business deliverables (research, decks, sites), Codex for editing an actual code repository.
01:20 – 02:03
03 · Setting up Dot
Dot is introduced as a persistent assistant with its own computer, available in the ChatGPT desktop app sidebar on the Pro plan. Grace names hers Emily, connects her to Slack, and sets up email and voice access.
02:03 – 02:34
04 · The AI marketing team being built
The target structure: one Marketing Team Lead at the top, five specialist agents underneath (Growth Strategist, Content Creator, Creative Designer, Website Designer, Growth Analyst), and Dot sitting outside the team as an executive assistant.
02:34 – 03:27
05 · Setting up the brand context
A local project folder holds a context file describing the brand (Lexora, an online language-learning platform) and an agents.md onboarding handbook. A free ChatGPT Marketing Team Setup Kit scaffolds both automatically.
03:27 – 05:03
06 · Creating the first marketing skill
Inside ChatGPT Work, Grace teaches an ad-creative skill in one prompt. ChatGPT saves it as a reusable skill file; tagging a campaign brief afterward triggers it to generate on-brand ad images automatically.
05:03 – 06:02
07 · HubSpot x OpenAI AI Growth Bundle (sponsor)
Sponsor segment for HubSpot's ChatGPT integration: a CRM connector that feeds real customer data into ChatGPT, plus a discounted bundle offer for new advertisers.
06:02 – 07:49
08 · Using Dot with an agent skill
A second skill, website building, gets created, then handed to Dot (Emily) directly instead of prompted inside the project. Emily locates the brand project, builds a one-page offer site, deploys it on ChatGPT Sites, and iterates on feedback.
07:49 – 08:27
09 · Skills vs. agents: when to use which
The rule of thumb: use a skill for repeatable, clearly-defined tasks where consistent output matters; use an agent when the task needs judgment, interpreting a goal, choosing skills, and owning a specific area of work.
08:27 – 09:51
10 · Creating the first marketing agent
A Growth Strategist agent is created with just a persona, focus, and scope. Tagged inside the project with a new goal, it automatically picks relevant skills and produces a strategy deck and campaign ideas.
09:51 – 12:34
11 · Using Dot to orchestrate the agent team
A Content Creator agent is added, then Grace messages Emily through Slack asking for two weeks of campaign social content. Emily confirms the brief, coordinates the agents, and returns a review PDF, six post drafts, and a full posting schedule about ten minutes later, then schedules approved posts onto a content calendar.
12:34 – 15:58
12 · The full AI team in action
A multi-agent lead-generation task: a ten-page guidebook, a landing page with a working lead-capture form, and a hosted site, built and deployed through ChatGPT Sites in about fifteen minutes. Grace closes with how to reach Dot by Slack, email, or voice, and a pitch for her paid AI Growth System program.
Lines worth screenshotting.
- ChatGPT Work handles multi-step business deliverables like research and reports, while Codex is the same underlying technology aimed at editing and inspecting an actual code repository.
- Dot is a persistent ChatGPT assistant with its own computer that can remote-access and control a user's files with permission, and it is currently limited to the ChatGPT Pro plan.
- A five-agent marketing team splits into a Growth Strategist, Content Creator, Creative Designer, Website Designer, and Growth Analyst, each reporting to one Marketing Team Lead.
- A persistent assistant can sit outside the project hierarchy entirely, taking tasks from anywhere rather than being locked to one single project.
- The brand setup is just a local project folder: a context file describing the brand and customers, plus an agents.md file that acts as an onboarding handbook for how the agents should work inside that project.
- The rule for choosing skill vs. agent: use a skill when the task is repeatable and needs the same quality and format every time; use an agent when the task needs judgment to interpret a goal and pick the right skills.
- A single prompt teaching ChatGPT Work an ad-creative skill creates a reusable skill file, callable afterward just by tagging the campaign brief and asking for ads.
- Messaging a persistent assistant inside Slack can trigger it to coordinate multiple specialist agents on a task, then report back status updates while the user is away from the desk.
- A single Slack message asking for two weeks of social content returned, about ten minutes later, a full campaign package: a review PDF, six post drafts, and a posting schedule, each post already organized into its own folder.
- The same assistant can push finished content straight into a connected tool, such as scheduling approved posts onto a content calendar.
- A multi-agent task that builds a ten-page guidebook, a landing page with a working lead-capture form, and a hosted site ran end-to-end from one prompt in about fifteen minutes.
- Once a brand's context file and agent/skill library exist, a user no longer has to re-explain the brand or give step-by-step instructions on later requests.
- The assistant is reachable outside the desktop app too, through Slack, email, or a voice call, so a task can be delegated the moment an idea comes up away from the computer.
Skills handle repetition; agents handle judgment.
THE DELEGATION RULE
The whole system works because of one rule: give ChatGPT a fixed skill for repeatable work and a judgment-making agent for everything else, then let a persistent assistant route between them.
02ChatGPT Work vs. Codex: where to build
- ChatGPT Work and Codex run on the same underlying model, but Work is built for broad multi-step business deliverables (research, decks, sites) while Codex is a developer environment for editing an actual code repository.
- The deciding question for which tool to open: if the deliverable is a business outcome, use Work; if the deliverable is software, use Codex.
03Setting up Dot
- Dot is a persistent ChatGPT assistant with its own computer, reachable from the desktop app sidebar, Slack, email, or voice, currently limited to the Pro plan.
- Dot can be renamed and personalized, and connecting it to Slack and email turns it into something closer to a colleague than a chat window.
04The AI marketing team being built
- A workable agent org chart is flat and small: one team lead, five specialists (growth, content, creative, website, analytics), each owning one area of work.
- A persistent assistant can sit outside that hierarchy entirely, working as an executive assistant with access to any project rather than being scoped to one.
05Setting up the brand context
- The entire brand setup is two files in a local folder: a context file describing the brand and customers, and an agents.md onboarding handbook explaining how the agents should work.
- A free starter kit can scaffold both files automatically, which removes the blank-page problem of defining a brand context from scratch.
06Creating the first marketing skill
- A reusable skill gets created from a single natural-language prompt describing the job; ChatGPT saves it as a file it can reuse without being re-taught.
- Once a skill exists, triggering it is as simple as tagging the relevant brief and asking for the output, no re-explaining the process each time.
08Using Dot with an agent skill
- A second skill handed to the persistent assistant directly, outside the project chat, still gets applied to the right project and brand without extra setup.
- The assistant can route a task to a different tool automatically and return a working, hosted first draft without the user manually switching tools.
09Skills vs. agents: when to use which
- Use a skill when the task is repeatable and the process is clear; the goal is the same quality and format every time.
- Use an agent when the task needs judgment: interpreting a goal, choosing which skills apply, and owning a defined area of work.
10Creating the first marketing agent
- An agent's setup can be minimal: a persona, a focus, and a scope. The value of an agent is the judgment layer on top of the skills it has access to, not more detailed instructions.
- Given a new goal, a well-scoped agent picks its own relevant skills rather than being walked through steps, and still produces a formatted deliverable.
11Using Dot to orchestrate the agent team
- A single Slack message with a plain-language goal is enough for a persistent assistant to pull in multiple agents and produce a complete deliverable set.
- The returned output was pre-organized: a review document, one folder per post, and a posting schedule, the kind of structure that normally costs manual setup time on a real campaign.
- Once content is approved, the same assistant can push it straight into a connected scheduling tool, closing the loop without a separate manual step.
12The full AI team in action
- A full lead-generation campaign, a guidebook, a landing page with a working lead-capture form, and a hosted deployment, ran from one prompt in about fifteen minutes.
- Past campaign performance data handed to the team was used to inform the new campaign, so the system compounds on what is fed into it over time.
- Once the brand context and the skills/agent library are set up, later requests stop requiring the brand to be re-explained or instructions to be spelled out step by step.
Terms worth knowing.
- ChatGPT Work
- OpenAI's workspace mode inside ChatGPT for multi-step business deliverables such as research, reports, presentations, spreadsheets, and websites.
- Codex
- A developer-focused ChatGPT environment for working directly with a code repository: inspecting, editing, and committing files.
- Dot
- A persistent, nameable ChatGPT assistant with its own computer, reachable from the desktop app, Slack, email, or voice, that can access and act across a user's projects. Currently limited to the ChatGPT Pro plan.
- Agent skill
- A reusable, taught process saved as a file that ChatGPT can trigger again on request, used for tasks where the output format and quality should stay consistent.
- Agent (ChatGPT Work)
- A persona-scoped sub-assistant with a defined focus and role, used for tasks that require judgment, such as interpreting a goal and choosing which skills to apply.
- ChatGPT Sites
- ChatGPT's built-in hosting and deployment feature for pages and sites it generates, including a lead-capture database, without needing an external server.
- agents.md
- A project file that acts as an onboarding handbook, telling the agents and skills in a project how they should work together.
Things they pointed at.
12:08toolNotion content calendar
Lines you could clip.
08:14
“Use the skills when the task is repeatable and the process is super clear. Use an agent when a task needs judgment.”
02:53
“This is just like an onboarding handbook for the agents, to understand how to work inside this project.”
11:10
“I can just walk away and let my assistant handle the rest. That's the biggest value I found in using Dot.”
12:08
“It's really like working with a real teammate.”
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.
analogy
ChatGPT has got so many upgrades recently, and it almost feels like a completely different app compared to just a few months ago, especially when it comes to building AI agents. Now it has ChatGPT Work, ChatGPT Codecs, and even a 24 -7 agent doc you can message on the go. So in this video, I will show you the best setup and how to create your own AI marketing team with ChatGPT.
No coding is required. Let's do it. Now I know it can be quite confusing since we have ChatGPT Work, Codex, and now Docs.
So ChatGPT Work and Codex use much of the same underlying Asian technology, but they're designed for different kinds of work. ChatGPT Work is where you dedicate broader, multi -step work. Research, analysis, reports, presentations, spreadsheets, websites.
It's designed around the finished outcome you want. Codex is more developer -focused environment. It's built for working directly with with code base, inspecting files, editing codes across the repository.
So for me, I will use Work when the deliverable is a business outcome and use Codex when the deliverables is a software or program. Now, Docs is a named assistant designed to take an ongoing responsibility. It has its own computer and can use Work, Codex, and the tools you choose to connect.
So they're not separated but can work together. For this video, I will use the ChatGPT desktop app. It lets me set my docs and work on my other projects.
projects from the same place note for now dot is only available for the chatgbt pro plan so once you download the chatgbt desktop app you should meet your dog at the sidebar so dot has its own computer and can also remote access and control your computer files with permission you can even customize the doc by giving it a name so I'm going to call her Emily and I can even add her to my existing Slack channel so I can message her anytime I can even set up an email for Emily as if she is my colleagues or call her and talk to her, which is powered by the GPT voice model.
And I must say the experience is pretty fun and amazing. And I'm going to show you how you can use it. Now for this video, we're building a five agent marketing team.
At the top, a marketing team lead, the main agent that I talked to. And under it, we have five marketing specialists, a growth strategist, content creator, creative designer, website designer, and a growth analyst. So each owns one area of work.
And my Emily, our dog, sits outside this team and works. as my executive assistant rather than being assigned to just one project. So I can give her a task from anywhere and she can still work on any project that she has access to.
So to start building an agent team, we will first set up a local project folder to keep everything organized. So in this case, we are working for a brand called Lexora, which is an online language learning platform for working professionals. So inside we have a folder to store the context file as well as other marketing output folders.
Basically, you just tell it what your brand does, your target customers, brand style, etc. And you can definitely include more than that. And then we also have this agents .md file.
So this is just like an onboarding handbook for the agents to understand how to work inside this project. But don't worry, I have this Charging Team Marketing Team Setup Kit. So you can download it and it will set up the basic context file and structure for you.
So you can grab it for free below. Then when everything is ready, we can create a new... projects on chatgbt and add back this folder as a source so now ai should know about your brand but that doesn't mean it knows how to work so that's why the first step is to create a reusable agent skills that teach it how to do things so now let's create our first agent skill so which is an ad creative skill and note we will stay inside chatgbt work and you don't need to switch to codex so suppose these skills will be through the brief from the user and generate ad concepts and images with this image generation models so pretty quickly it has already created our first skills and if we navigate to the project folders we will see it has automatically created a new skill under this dot agent folder and again do not worry if you don't understand that you can just go to the skill panel and you will see there is an add creative skill created for this brand project no coding is required so now whenever we need creative ads we can just hand it to
So for example, just tag the campaign brief file and then ask it to generate the creative ads. So you can see it is now calling the ad creative skills we just created and then generate the visual concept for the Instagram platform and then adding the brand logo and copy to create a final ads. And here is the results.
So you can see they have a lighter visual style consistency and also matching the branding with professional tone. Also using the brand color, both the CT buttons, logo treatment. generated image access, and everything have been saved locally, including the ad concept so you can share back to the team if you want.
And so the skills will give ChatGPT a reusable process to follow next time. Now we're building an AI marketing team to do it faster. But the way people discover business is changing too.
Your customers are now asking ChatGPT for recommendations. So how does your brand show up there? Hotspot, a longtime partner for this channel, just launched something new with OpenAI called the AI Growth Bundle.
HubSpot is the first CLM plugged straight into ChatGPT. So you can create ChatGPT ads and see results alongside your other campaigns. And they just released a new smart CLM that actually updates itself.
So the AI note -taker records every call, saves the transcript, and drives your follow -up automatically. And what I found the most useful is that the HubSpot connectors bring your actual CLM data into ChatGPT because your business context is the most valuable thing you can fit in. They're offering a significant discount on HubSpot started plus buy one get one on chat gpt businesses for a year and up to 500 in master chat gpt s band for eligible new advertisers so get started and claim your offer through the link below and thank you hotspot for sponsoring this video just now we created our first agent skill but we can create more than that so it gives your marketing agents different playbook to do the work and this time we're also gonna give emily our dogs to use these marketing skills so this time let's create a website builder scale so this scale we create create high -impact websites and landing pages based on our marketing goals.
So this time, we will also ask it to create a design source file that contains a list of great design sites for the agents to take reference. And pretty quickly, we have created second skills for this agent team. And this time, instead of prompting inside its project, I can just message Emily, our assistant, to locate the brand project and then help to build a one -page offer website for the upcoming free trial campaign.
And then I can also specify and ask it to use ChatGPT sites for the deployment. And what she's really doing is routing my tasks to Codex and initiating a new thread. So it has given me the first draft version.
And then you can see I'm iterating back and forth with Emily. And here is the final result, which is now hosted on ChatGPT and is ready to share with the team. And I like the overall style.
And also it is interactive where I can click on certain web element and then drag within the server. not too aggressive and the layout also looks professional with clear edges lines which i like it and all the images are generated using the latest gbt image model and i would say this is pretty impressive and it just took me less than 30 minutes and of course you can always enhance it so for example maybe you can add your own proprietary images or even videos and the great thing is now with this website design skills we can create more web pages and landing pages as we need them and your agent or docs can just follow the same process and approach so using similar methods we can build more skills to a total 50 marketing skills library from strategy content creation to website building but at a certain point you want to start creating specialized agent so when do you just use the skills and when do you actually need an agent the answer is really simple use the skills when the task is repeatable and the process is super clear you want the same quality and the same format every time like
building a page or creating an app use an agent when a task needs judgment it has to interpret a goal design what matters pick the right skills and own a specific area of work and to give you a better idea let me show you how you can create your first marketing agent inside chat gpt in this case we're creating a growth strategies agent so these days i try to be really simple when creating agents and just to give the persona focus scope and the role because the biggest value of the agent is the judgment so this strategist will turn a goal into a strategy document using research or data and quickly we have created the first agent and if you navigate to the project folder again you will see chatgpt create a folder doc codex which contains the agent file inside again you don't need to touch any coding files chatgpt will do that for you now whenever you want to ask this agent to do the task you can just tag it right inside this project so for example we can
give it a new growth goal and ask it to propose new campaign ideas or strategies and you can see it will automatically pick these relevant skills and that's why you're letting the agent to decide that for you rather than giving it step -by -step instructions and then it will comes up with the campaign idea and also the strategy deck and i like how the deck is being pulled together it professionally done every slide has a core idea and i found the extra model is really good at making this type of diagrams and flowcharts And I intentionally don't want too much fancy design to match the branding.
And it just did that for me. And of course, you can also ask it to include images throughout the slide deck. So that's one specialist agent.
And it already saves you hours. But I'm still prompting at my desk. So this time, I'm going to show you how we can push it further.
So this time, I will message our docs through Slack and have it coordinate with our agent team. So first, create our second agent, which is a content creator. And again, we can just follow the same process.
process give it the role and persona so these content created agent will create content things in the audience and produce content like social poses blog articles etc and then pretty quickly we have created our second marketing agent for this team so just now you know we can tag this agent right inside this project but this time let's try to message this agent through slack so remember we have already add Emily or your own dog to slack so let's say the team has approved the campaign brief we just And then now I can just tell Emily to coordinate that with the team and say we need two weeks of social content for the campaign launch.
And you can see she immediately replied me back and saying she will find the confirmed brief, get the team's input and send me back the schedule. And she even pinged me back to keep me updated on the status while the team is working. So imagine I can just walk away and let Emily to handle the rest.
And this is the biggest value I found in using Docs. to chat gbt you can see basically what emily did is route my request to my project and invoke the agents and in this case she has called the growth strategies and content creator agent we just created and then after like 10 minutes emily came back with the results and i would say the results is pretty impressive as well so it has compiled everything into this review pdf containing all six post draft so it's just like an overview briefing for the team about both the and copies and i can actually download all the content access from this zip file so you can see it has everything that i need the brief document the posting schedule and every post has its own folder so everything is organized for the campaign launch and to be honest i am pretty impressed because oftentimes when we do a campaign we are not just creating one separate access but a lot of different kinds of assets so having this folder structure is super helpful and everything is ready to post and now
can push it even further and ask emily to schedule the content on our notion content calendar and pretty quickly emily has got by so she has already scheduled all the poses using the content scheduling skills and everything is now ready on our content calendar for the team's assets and this is super handy once you have connected your docs to your tools and this is a use case i highly recommend you try the experience it's really like working with a real teammate so for this project we can continue creating more specialist agent and a marketing skills library they can use.
Of course, you don't have to follow this same hierarchy. You can even just start with one to two AI agents. But there is one more important step, which is to update the project instruction.
So if you remember, we have this agent .md file. Now we can ask ChatGPT to reveal the project agents and skills and then add a brief routing guidance so it has a better understanding of when it should dedicate to a specialist agent versus this using skills directly so now everything is set and let's give it a multi -agent task so this time let's test how the team works together right inside the project so go back to the brand project so let's say this time the team also wants to create a lead mana campaign in order to get more leads for the team campaign we just confirmed in q4 and this time i will also give it some past campaign performance and then ask it to use what we have learned to inform these new lead mana So you can see it's giving me its plan.
What are the skills that they plan to use and confirm me that the analyst is already reviewing the results. So we can actually click on the sub -agents tab and you can see this is your multi -agent team working in action. And after like 15 minutes, the team has ready the outputs.
So let's first review the lead manner. So it is a team meeting guidebook and I like how it is professionally done like the website with on -brand images, layout, practical tips. tips worksheets to make it highly valuable and end with a final CTA and then the campaign landing page which it hosted locally for testing first it is also nicely done so with the same guidebook cover it just created as well as the inner pages and this is completely handled by chatgpt automatically so this is pretty impressive as well and also the lead form embedded where we can try testing the lead capture flow so if you use any external lead capture tool you can just give chatgpt the integration or maybe the form id and let it handle it so you can see everything is working great it will give me back the actual guidebook pdf after filling out the form and once you're happy with it i can even ask it to host and deploy using the chat gpt sites so where it has its own built -in database for the leak capturing and you don't even need an external server everything is hosted using chat gpt site bonus tips you can also use doc for this kind of multi -agent task and you can always
message it through Slack, just like what I did. And you can even email it with a task or even call it and talk it through. So if an idea comes out while you're away from a desk, you can just let it handle that for you.
So you would notice that once we set up a system like this, you don't even need to explain the brand again or give it step -by -step instructions. It will automatically know what to do next. So if you want to build this for your own brand or team, also check out our eight modules AI growth system for marketing program.
It walks you through building your own ai system step by step rather than just collecting more ai tools so when you join the program you will also get access to our community asking questions access our past workshop prime resources and templates you can use i hope to see you there and before you go also watch this whole video playlist about how to better use chat gpt for your marketing i will see you next time
The Hook
The bait, then the rug-pull.
Grace Leung opens by naming the problem: ChatGPT has shipped Work, Codex, Skills, Agents, and now Dot in quick succession, and it's genuinely confusing which one does what. She spends the next sixteen minutes building an actual five-agent marketing team end to end, so the differences stop being abstract.
Named ideas worth stealing.
02:03model
The Five-Agent Marketing Team
- Marketing Team Lead
- Growth Strategist
- Content Creator
- Creative Designer
- Website Designer
- Growth Analyst
A simple org chart for a ChatGPT-native marketing team: one lead agent tagged for day-to-day work, five specialists underneath each owning one area, plus a persistent assistant (Dot) sitting outside the hierarchy with access to any project.
Steal forany one-person team automating repeatable marketing work
08:14concept
Skill vs. Agent Decision Rule
Use a skill when the task is repeatable and the output quality/format needs to match every time. Use an agent when the task requires judgment: interpreting a goal, picking the right skills, and owning a defined area of work.
Steal fordeciding what to automate as a fixed process vs. what to delegate as a role
How they asked for the click.
VERBAL ASK
15:03product
“If you want to build this for your own brand or team, also check out our eight modules AI growth system for marketing program.”
Soft pitch after demonstrating full value, pointed at her own paid program (grooveanchor.com), plus a plug for the free setup kit and a related video playlist.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
