Big Idea

The argument in one line.

OpenAI Codex's /goal command unlocks days-long autonomous agent runs for outbound, recruiting, and ops at a fraction of the API cost — turning AI from a chat toy into a background revenue engine.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run a services business or agency and want AI agents handling outbound prospecting without reviewing every step.
You are paying $500+ per month in AI API costs and looking for a more generous-limit alternative for long-running tasks.
You have or want a team and need a way to systematize processes into reusable agent skills everyone can run.
You are interested in practical, real-company applications of AI agents beyond demos and toy examples.

SKIP IF…

You want polished, edited tutorials with clear step-by-step instructions — this is a live, somewhat scattered screen-share.
You are looking for consumer use cases; this is squarely focused on B2B sales, recruiting, and agency ops.

TL;DR

The full version, fast.

Codex can run business tasks autonomously for days using the /goal command, with usage limits generous enough to avoid the $12,000/month API bills that come with sustained Claude use. The video shows six live examples: personalized ad creative at scale tied to cold outreach, an autonomous recruiting pipeline that daisy-chains LinkedIn, email, and enrichment tools, an internal skills marketplace for company process consistency, a screen-monitoring agent that auto-detects which workflows to systematize, and a Dream 100 command center for revenue action visibility. The underlying playbook is the same for all six: solve a business bottleneck with a Codex agent, then optionally sell that solution to others.

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.

00:00 – 00:47

01 · Hook — Codex as revenue engine

Opens with the claim that Codex is generating real revenue, frames it as overlooked while everyone debates Opus 4.8.

00:47 – 02:48

02 · Use Case 1: Picasso to Outbound

Ad creative agent that forks a GitHub repo and generates personalized ads per prospect for cold outreach. Has been running 28 hours.

02:48 – 03:20

03 · Bonus: Hermes + usage limits

Credits Hermes as more reliable than Claude Pro for sustained multi-day runs. Notes generous usage limits vs Claude Pro.

03:20 – 05:31

04 · Use Case 3: LinkedIn Recruiting Pipeline

72-hour running goal that scrapes candidates, enriches them, sends personalized cold emails, and DMs via LinkedIn. Daisy-chains multiple sourcing tools.

05:31 – 08:04

05 · Use Case 4: Skills Dojo

Internal skills marketplace where company processes are documented agent skills. Tracked by downloads, runs, upvotes, forks, users.

08:04 – 08:25

06 · Sponsor — Karrot

Personalized LinkedIn ad landing pages. karrot.ai.

08:25 – 11:09

07 · Use Case 5: TerraMind Process Skill Monitor

Screen-watching agent that identifies which workflows to skillify. Demo shows it flagging Slack-to-Hermes, outbound approval autonomy, and revenue lifecycle as top leverage.

11:09 – 12:36

08 · Use Case 6: Dream 100 Command Center

Read-only dashboard showing which of the top 100 target accounts need a revenue action — pulling from Carrot, ClickFlow, sponsorships, and future lanes.

12:36 – 12:56

09 · Wrap — solve your own problems first

Closes with the thesis that solving your own problems with Codex agents is the starting point, and everything can be sold or given away to get customers.

Atomic Insights

Lines worth screenshotting.

Codex's /goal command runs agent tasks for days without human intervention — making it the right tool for outbound and recruiting automation, not just code generation.
Personalized ad creative per prospect at scale is now a solved problem: fork a GitHub creative repo and let an agent customize 100 ads in an outbound run.
Daisy-chaining recruiting tools (sourcing agent + LinkedIn Recruiter + Instantly cold email + personal email finder) with one Codex goal beats any single recruiting SaaS.
Internal skills marketplaces — where every company process is a documented, runnable agent skill — are the new process documentation.
A screen-monitoring agent that watches what you do and recommends which workflows to skillify removes the discovery step from process systematization.
Whatever problem you solve for your own business with a Codex agent, you can immediately productize and sell to others facing the same problem.
Claude Pro account usage limits run out fast on sustained agentic tasks; Codex is materially more generous for the same workload.
A Dream 100 command center gives one person visibility across every revenue action across every platform without checking six different tools.
The bottleneck in most outbound operations is not ideas or contacts — it is messy context that never resolves into a decision or sent message.
Spending $12,000 per month on personal API tokens is the real cost of using Claude's API for sustained agent work — Codex eliminates that category of spend.

Takeaway

Six ways to run AI agents that actually work.

WHAT TO LEARN

The value of an AI agent is not what it can do in one session — it is whether you can set a goal, walk away, and come back to revenue-moving results.

The /goal command in Codex enables multi-day autonomous runs — outbound, recruiting, and ops tasks can run for 72+ hours without your direct involvement.
Personalized outreach at scale is solved: fork a creative asset repo, let an agent generate custom variations per prospect, and tie it directly into your cold email sequence.
Daisy-chaining multiple sourcing tools through one agent goal beats any single recruiting SaaS because you own the logic and are not locked into one platform's limits.
An internal skills marketplace turns your best processes into reusable, trackable agent skills — the same way a good SOP library works, but each skill is actually executable.
A screen-monitoring agent that identifies what you do repeatedly and recommends which workflows to automate removes the hardest step in process systematization: knowing what to systematize.
A Dream 100 command center gives one person full visibility into which target accounts need a revenue action without logging into six platforms.
Whatever agent you build to solve your own problem can be sold or given away to others with the same problem — the build-to-sell pipeline starts with solving for yourself.
Sustained agentic workloads have a real API cost ceiling; knowing the usage economics of each tool determines which platform is right for which job.

Resources

Things they pointed at.

00:00toolCodex (OpenAI) ↗

02:35toolHermes autonomous agent

01:40toolClickFlow MCP

09:05toolTerraMind

03:30toolLinkedIn Recruiter ↗

03:45toolInstantly ↗

12:06toolApollo ↗

12:06toolClay ↗

08:04productKarrot ↗

Quotables

Lines you could clip.

00:00

“Codex has been insane in terms of generating revenue, and I'm gonna show you six crazy ways where it is actually making money for me and my business.”

Strong revenue hook with a numbered promise→ IG reel cold open↗ Tweet quote

05:29

“I'll tell you, spending up to $12,000 a month on my personal API tokens, that's not a good thing.”

Specific dollar figure + pain point relatable to any AI-heavy operator→ TikTok hook↗ Tweet quote

06:37

“Skills, to me, are a form of having good process.”

Reframes AI skills as process infrastructure — quotable insight with no context needed→ newsletter pull-quote↗ Tweet quote

07:20

“Whatever problems you solve for yourself, you can probably go out there and sell this.”

Universal, standalone business lesson→ IG reel cold open↗ 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

00:00Codex has been insane in terms of generating revenue, and I'm gonna show you six crazy ways where it is actually making money for me and my business. Now, look, I get it. Everyone's talking about Opus four eight right now, but I'm telling you, you gotta pay attention to Codex because the things that it can do right now, it's absolutely crazy.

00:18So let's get into it. So the first thing I wanna show you is that we actually have an agent called Picasso that can make ad creative at scale. So this is an example of ad creative that we might make, and you can see that it looks pretty damn good, and this is a set of 100.

00:30Now, what I was thinking is that this is sitting in our GitHub repo, and I want to have the ability to fork this repo and then tie it into what we do from an outbound sales standpoint. And so if you think about it this way. If you're able to basically you're able to make ads, custom ads for every single prospect that you send Outreach to, you're gonna stand out way more than anybody else.

00:51And it's high quality ads like this, that looks good. Right? And so what I wanna bring your attention to now is if we take a look at my codex, you can see it's actually working right now.

01:00So inside of my codex, if you look over here so you can see it's called Picasso to Outbound over here. Okay? So by the way, it it's been working for the last twenty eight hours in terms of forking this over, and it looks like it's it it should be done soon.

01:13But I all I did was I took the fork of the I I took the GitHub repo and I forked it, which means that I just splintered it off to what I'm trying to do from an outbound standpoint. And so think of it, if you have a an ICP, an ideal client profile that you're looking to reach out to, maybe you have a dream 100, these are the top 100 accounts that you're trying to win, well, you can do that now.

01:33So not only can we do this with Picasso, but we can also use the ClickFlow MCP, which is the AEO SEO software that we have, which generates high quality content. And we can use it saying, hey, we did research.

01:44We know what your angles are. We know who your competitors are. We think you should be making this content over here.

01:48Here's five examples. Would you like us to do more? Right?

01:51You can do the same thing with creative. You do 10 examples, you don't have to do a 100. Hey, here's an example.

01:55Would you like us to do more? And if you're able to do that, that's gonna generate more pipeline for you, and it's ultimately gonna grow your company a lot faster. And so that's how you you wanna think about doing stuff that's outside the box.

02:05And so the cool thing is, my amazing team, Chris and Eugenio, they they they built this out, and I'm just like, okay. Well, if you you're very active, and you have a team helping you already, and and you can fork this stuff, amazing.

02:17Or if you can find other stuff on GitHub to fork, that's gonna work out well for you too. And the cool thing is when you use the slash goal command inside of Codex, you can do this with with other tools too, but I just find that using using 5.5 and then using Codex has been great, and then even my my Hermes, my autonomous agents have been great for that as well.

02:34So I do wanna give credit. Number two would be a little bonus here, giving credit to using Codex, not just the app, but the way that you can use it with Hermes, which is their autonomous agent, that is extremely powerful too. I I found it generally to be more reliable than using using Cloud for it.

02:50So and bear in mind, I I I pay for both. So I'm gonna let this continue to work over here, but again, you can you can let this work for I've seen people let this work for days at a time, and it's it's been it's been really good.

03:00And I do also wanna call a little a little bonus here. If I look at the the the usage, it it's actually pretty good. You can see the usage over here.

03:08It's they're very generous with their usage usage limits, whereas when you're using Claude, you can actually run through them very quickly. So just keep that in mind. Alright.

03:16So number three, check out this goal over here. I've been working on this for the last seventy two hours or so. So for my company, for Single Brain, we're trying to recruit.

03:24We're trying to recruit engineers. We're trying to recruit customer success people. We're trying to recruit really smart marketers.

03:28So we are in a recruiting binge right now, and we pay for a handful of tools. And so I'll give you an example. We pay for one tool right now that basically is a sourcing agent for us.

03:37Okay? It finds people's personal emails, and it sends them their emails. Okay?

03:40And it's they're they're pretty generous with with their with their limits. Now, the other thing I'll say too is that we use LinkedIn. We pay for LinkedIn recruiter.

03:47We also pay for instantly as well, and we have inboxes that we use for sending cold emails so we can send more volume. Right? We have 10 inboxes just for sending out cold emails when it comes to recruiting, and we haven't been using those.

03:58And so think about it this way. Instead of just using that one recruiting agent that's just putting people into sequences, what if you can take all the contacts that they found, send them cold emails, customize each one, okay, not just using the sequences because you wanna customize these.

04:13So they're cuss each one is customized. Right? That's one piece.

04:16But then the other piece is you're also combining it with Outreach from your LinkedIn. So you're DMing and you're adding these people to your your profile as well so they can see the content that you're continuing to put out there.

04:27Not only that, you can also tie so like I just mentioned earlier, if we're using instantly to send out emails, you can tie it in with that. So you don't you're not tied into one platform. And so if you can daisy chain a couple of these things when it comes to a recruiting standpoint, from a recruiting standpoint, you're gonna stand out versus other people because it's you're taking a multi threaded approach.

04:43And so for me, this matters quite a bit because if you generate all these sales if you generate all the sales in the world, but if you don't do a good job with the recruiting, that's gonna be an issue. Right?

04:51You gotta bring in the best people. But not only that, what if you build something like this and then you sell it? You can sell this recruiting agent to people, and people are gonna find that solution is is very, very good.

05:01Right? And so you you can see in this case, these things continue to work over here, and you can see, okay, we're we're testing this right now.

05:07So get the get the get the next 10 qualified SEO paid media recruits that we should be having in here. And this will continue to work.

05:13You can see what's happening over here. And occasionally, it's gonna ask me to, you know, review people over here. I review it, and then it gets back to work.

05:19Okay? So that's why I I like the slash goal command so much. I think this is really important.

05:23Codecs is really important because their usage limits, again, are very generous versus when you use a Claude when you use Claude's pro account, it's it's not that generous. Know you're gonna have to switch to API.

05:32And I'll tell you, spending up to $12,000 a month on my personal API tokens, that's not a good thing. That's not something that you wanna do. Okay?

05:39So I would say this is this is number three on the list. Right? But not only that.

05:44I would say if you wanna generate more revenue, you gotta arm the the you gotta arm your team. Okay? So you just you just talk about arming the rebels over here.

05:51So number four, we've created a skills dojo, or we call it the skills academy. Okay?

05:55Imagine you have all these skills inside of your your company. Okay? Everyone has their own skills.

06:00You might have different, you know, sales skills, you might have skills that will analyze, you know, your traffic going down, or you might have a skill that's AEOSU opportunity finder, you might have one for, you know, you might have one for reviving deals as well.

06:16And the cool thing is this, you can also sort it by you can sort it by the different categories as well if you want to, but the cool thing is you can see how many of these skills have been downloaded, how many have ran, how many upvotes there are, how many forks there are, how many users there are for these. The whole idea is that you want your company to have more consistency when it comes to when it comes to how you do things.

06:38Right? Because people all all oftentimes will talk about if you wanna scale a company, you have to have good process. Well, skills, to me, are a form of having good process, and what better way to follow a process than to have all these documented and have a, basically, a skills marketplace that your team can go to and use this.

06:53Right? And imagine if your sales team is using all the skills and they're all marching in the same direction, fulfillment's marching in the same direction, what's gonna happen? You're gonna have more consistent execution, which will then lead to higher conversion rates, which would then lead to more revenue for your business, you're gonna make more money.

07:09Okay? The cool thing is you can you can submit this as well. So just keep this in mind.

07:13Like, everything that you're trying to do ultimately, you can do it for yourself, and then you can figure out, like, whatever problems you solve for yourself, you can probably go out there and sell this. So I don't care what level you're at. Maybe you you have a business right now, maybe you don't have a business.

07:23All this stuff, if you learn how to do this, is gonna make you more money ultimately. And and, you know, with this one, I specifically, I I've gotten it to a point where I've I've I've pushed it back to GitHub, and now my team is gonna take it from there.

07:36They're they're they're gonna there's there's a handoff component. Right? But don't don't be like, oh, just because because because I have a team doesn't mean that you can't do this.

07:42You definitely can. If you can learn how to do this on your own, eventually, you're gonna be in demand. Other people are gonna want you to do it, what's gonna happen is you're gonna have to hire people.

07:50Right? So it's it's all gonna move in that direction. K?

07:52Quick break. If you wanna run personalized LinkedIn ads and have personalized landing pages to convert your customers at a much higher rate on LinkedIn, check out Karrot. That's karrot.ai.

08:04Karrot allows you to do things that LinkedIn ads does not allow you to do right now. Check it out. There are publicly traded companies such as Semrush and Sitecore that are using this right now, and you can use this to get ahead.

08:14Because if you don't use this, it will take you weeks, if not even a little longer than that, to make these ads. Because again, you cannot do this on LinkedIn.

08:22So again, go to www.karrot.ai, and we'll see you on the other side. And then number five, what I've also done is I have created a the ability to this thing's called TerraMind.

08:34Okay? So what TerraMind does is it is actively looking at what I'm doing on my screen. Okay?

08:41So it's it's basically surveilling, and I don't wanna use this as a negative word, but surveilling what I'm working on all the time, and then it's it's figuring out what we should make into a skill. Okay?

08:51So it says, yes, but I'd ruthless. Only three are truly high leverage for you right now. So Slack to Hermes.

08:56I use Hermes a lot inside of Slack. That's my autonomous agent. And plus a command queue.

09:01This is the highest leverage because your bottleneck is not ideas. It is messy Slack context turning into half decisions. I make Hermes send Teramind one so Teramind's the the software that we're looking at, and then we're we're looking to iterate off of that.

09:12And Teramind is it basically will will help your team Skillify it. Like, it'll observe what they're doing and Skillify it. I think that's what they do.

09:19And so I use Codex to build something out like that. Right? So owner deadline, next action, draft approval needed, blocked.

09:25This becomes your daily do do this, revealing this queue. Okay? That's pretty cool.

09:29Number two, outbound approval autonomy. This is high leverage because you clearly want outbound on scale without you reviewing every packet.

09:35So right now, because we're building all these things and I'm reviewing it myself, it picked this up, and so that's pretty cool. Like, it's doing these snapshots. Right?

09:42And so right now, it's it's not very intrusive at all. And imagine if you could roll this out for yourself. Imagine you could roll this out to your team to say, hey, like, all the things we're trying to take the things that you shouldn't be doing off your plate.

09:53We should skillify them. Maybe you can find it in a skill marketplace, and then we can scale out the organization. Right?

09:58So that's how this works. And then you can see I'm continuing to work with the back and forth. And look, I'm I'm doing another slash goal here, which I'll come back to in a moment.

10:05But having the ability to have a process skill monitor is amazing. Doing pick calls through outbound is amazing.

10:13The Skilled Dojo is amazing as well. I think the last one I wanna share right now, there there's there's more stuff for me to share, but the idea of of building a dream 100 command center. Okay?

10:22And what this means and by way, I was explaining it to somebody else earlier. I just asked to explain it in one sentence. So this Dream 100, earlier I talked about the top 100 people that you want to be constantly reaching out to or constantly nurturing.

10:34And if you win these 100 accounts, then you can build a really big business. So this Dream 100, it gives Eric one read only command center to see what revenue actions are ready, what approval what needs approval, what was sent or got replies, what was blocked, and what's next best money moving action that is across Carrot, which is a software that we have, ClickFlow, sponsorships, Dream 100, and future revenue lanes.

10:56Okay? So that's great. And so what I'm gonna do is this.

11:00I'm just gonna run this slash go over here because it's it's said these things were blocked over here. So I'm I'm gonna show you this in action, the request I hand off and back. Okay.

11:07Great. Great. And then we could use we don't even see the same values.

11:10Okay. Great. So I'm just gonna run this.

11:12Boom. And we're gonna replace this right here, and we're just gonna let it run. Okay?

11:16So keep in mind, you get very generous usage limits. And I what what I would say is if you run out, by the way, if you need to buy more, if you're using Hermes as an example, you you can buy multiple ones and just switch out the the OAuth. You can't really do it for this one because you you're in you're in one account.

11:30Just keep that in mind. Okay? And maybe you might be able to buy more usage out of this recording, maybe not, but that's what it is.

11:35So, anyway, those are a couple things. I mean, I will also say there's a handful like, I'll I'll give you a little bonus here. In terms of assessing cold email lead quality, there's another thing over here.

11:44So there's a handful of things that we pay for, tools that we pay for to generate leads. I've actually created my own, so I don't need to pay for something like Apollo or Clay, which I think those tools are great. I'm still okay paying for them, but maybe I want something that would generate the bulk of the leads, maybe like 80% or so at much cheap at a much cheaper rate.

12:00And then I can use these other tools to maybe find, you know, ultra high quality leads for maybe the most valuable ones I'm looking for, maybe the dream 100, and just keep going through that. So you can be a lot more strategic about how you do this.

12:12And because you can find any leads that you want, you're able to you're able to build your lead list more, and you're able to verify faster. And I also have something that will will verify leads as well by combining a handful of different tools that we have. And so that's what I would say.

12:26Ultimately, you wanna solve your own problems first, but you can see everything that I'm talking about here. I can either run for myself, or I can run for customers, um, or you can give these things away for free and try to get customers in.

12:37But it all comes down to what your relationship is with learning and what your relationship is with change. And so I would encourage you to learn as much as you can, and then you'll be in a really good spot. So hope you enjoy this video.

12:49We'll catch you in the next one.

The Hook

The bait, then the rug-pull.

The title promises six use cases. What the video actually delivers is a live window into a company running AI agents in the background 24/7 — and a quiet argument that the tool everyone overlooked while arguing about Claude vs GPT-4 is the one that is actually doing the work.

Frameworks