Big Idea

The argument in one line.

A non-technical founder can run a multi-million-dollar agency at five-person headcount by replacing SaaS workflow tools with a single custom-coded internal app maintained entirely by AI agents.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You run an agency or service business paying for a dozen SaaS tools to stitch together workflows.
You are not a coder but have considered building your own internal tooling with AI assistance.
You use n8n, Zapier, or Make and suspect agents cannot reason as well over those workflows as over real code.
You want a real cost breakdown of running Codex plus Devin as a production maintenance stack.

SKIP IF…

You want a technical deep-dive into the agent architecture -- this is executive-level, not engineering-level.
You are looking for a step-by-step tutorial on prompt engineering or writing code with AI.

TL;DR

The full version, fast.

Taylor Haren runs a $200k/month outbound agency with 5 employees by scrapping SaaS automation stacks in favor of a single custom internal app built and maintained entirely by AI agents. The core argument: agents reason better over real code than over Zapier or n8n workflow nodes, and no-code platforms are constrained to use the cheapest models to preserve margin. His production stack -- Codex as primary driver at $200/month plus Devin for PR review loops at $760/month -- keeps the entire codebase maintained for under a grand. The onboarding workflow that used to take hours fires automatically the moment a contract is signed, with AI handling CRM updates, Stripe subscriptions, Slack channels, Linear projects, and cold email provisioning end to end.

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 – 01:35

01 · Hook + thesis

Non-coder intro, revenue claim, custom internal app as the foundation

01:35 – 02:56

02 · Why custom code beats no-code

SaaS tools use cheapest models; agents reason better over real code

02:56 – 05:47

03 · Pre-onboarding workflow

Sales call triggers contract, billing portal, ideation form, launch call booking

05:47 – 07:00

04 · Full onboarding automation

Contract signature triggers CRM, Stripe, Slack channels, Linear projects, Instantly provisioning

07:00 – 09:06

05 · Background automations and Sentry loop

Devin monitors Sentry errors, diagnoses, writes PRs automatically

09:06 – 11:09

06 · Devin PR review loop

Auto-fix cycle: write code, review, fix conflicts, repeat until merge-ready

11:09 – 13:35

07 · Day-to-day setup

Codex as primary driver, Claude for design tasks, Vercel DeepSec for security, Mac mini plus mobile

13:35 – 14:14

08 · Cost breakdown and CTA

Under $1k/month total stack cost; 5 employees vs competitors 30

Atomic Insights

Lines worth screenshotting.

No-code automation platforms are structurally forced to use the cheapest AI models -- there is no margin otherwise.
Agents reason better over your own codebase than over Zapier or n8n workflow nodes because there is no API translation layer limiting context.
A non-coder can maintain a production codebase at $200k/month agency scale for under $1000/month using Codex plus Devin.
The biggest benefit of automating onboarding is not speed -- it is standardization: every client gets exactly the same experience.
Background automations triggered by events eliminate the human bottleneck that causes inconsistent client outcomes.
Devins PR review loop -- write, review, fix conflicts, re-review, repeat until clean -- lets a non-coder merge code they cannot read.
Logging every agent action in Linear creates persistent system memory any future agent can reference without relying on a human to remember.
A company at the same revenue running SaaS tooling and hiring for coverage needs 30 employees; this stack needs 5.
The custom billing portal built in a few Codex conversations replaced a Stripe invoice flow that could not handle conditional fee logic.
Devin discovered Instantly has a hidden front-end API and immediately used it to replace the browser-automation approach it had just built.

Takeaway

Why agents work better inside your own code

WHAT TO LEARN

No-code automation platforms are constrained to use the cheapest available AI models -- the only way to give agents full reasoning power over your operations is to own the code yourself.

Agents reason better over real code than over Zapier or n8n nodes because there is no API translation layer limiting the context they can access.
The Codex and Devin loop -- write new code, run the PR review cycle, fix conflicts, repeat until clean, then merge -- lets a non-technical operator ship and maintain production software without reading a line.
Background automations triggered by system events (contract signed, Sentry error fired) remove the human bottleneck that causes inconsistent client delivery.
Logging every agent action in a shared project management tool creates persistent system memory any future agent or team member can read without relying on anyone to remember.
Total AI infrastructure cost under one thousand dollars per month is achievable at multi-million-dollar service business scale -- significantly less than the SaaS stack it replaces plus the headcount required to run it.

Glossary

Terms worth knowing.

Devin AI: An autonomous AI coding agent that can browse the web, read codebases, write code, and open pull requests. Used here primarily for PR review loops and automated error remediation.
Codex (OpenAI): OpenAIs coding-focused agent product, used here as the primary driver for writing new features and code changes at $200/month.
Sentry: An error monitoring platform that captures runtime exceptions and sends alerts. Used here to route errors directly to Devin for automated diagnosis and fixes.
Linear: A project management tool used here both for human task tracking and as a persistent memory store that agents can read to understand system history and decisions.
Attio: A CRM platform used to store client contract data, automatically populated during onboarding by the AI-driven workflow.
Instantly: A cold email sequencer platform whose workspace setup is provisioned automatically during client onboarding via its hidden front-end API.
PR review loop: An automated cycle where Devin writes code, review agents check it, Devin fixes any flagged issues, and the cycle repeats until all reviewers approve -- requiring only a final human merge click.
Vibe coding: Colloquial term for building software by directing AI coding agents in natural language without writing or reading the underlying code yourself.

Resources

Things they pointed at.

00:40toolClaude Code

00:40toolCursor

00:40toolCodex (OpenAI)

00:40toolDevin AI

00:46tooln8n

00:46toolZapier

00:46toolMake

02:56toolAttio CRM

03:20toolStripe

03:50toolLinear

05:50toolInstantly

07:00toolSentry

11:40toolVercel DeepSec

00:07linkSales Automation Systems ↗

Quotables

Lines you could clip.

00:10

“I dont actually know how to write a single line of code. I dont even know how to read a single line of code.”

Credibility-building contrast -- massive claim delivered before the confession→ TikTok hook↗ Tweet quote

01:35

“The agents cant reason with those workflows the same way they can with your own code.”

The core thesis in one sentence -- quotable standalone argument→ IG reel cold open↗ Tweet quote

09:40

“I dont even know what the hell it says. I dont know what it does, but it works.”

Absurdist honesty that lands as both funny and validating for non-coders→ TikTok hook↗ Tweet quote

13:29

“For less than a grand a month as a multi-million dollar company, were able to maintain all of this.”

Concrete cost claim that challenges every founder paying more for SaaS→ newsletter pull-quote↗ 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.

00:00I run an outbound marketing agency that's doing over $200 a bot, and I do it with a extremely tiny team. Because almost everything inside of our operations from onboarding a new client the moment they sign up, to addressing angry client messages, to maintaining dozens of automations that keep us running, is now completely handled by Claude code and a stack of AI agents that talk to each other.

00:23And the part that still blows my mind is that I don't actually know how to write a single line of code. I don't even know how to read a single line of code. And so in this video, I wanted to just show fully transparently how Claude Code and AI agents are helping me run my business day to day, a little bit about how I have kind of built maintain them, maybe even some things that you can take away and implement for your own company.

00:45So the core app underneath everything in our own business is just a custom internal app that we've been using, and it's been completely built by Claude Code. Technically speaking, I started out on Cursor six months ago, then I moved over to Claude Code, like, three months ago, and most recently, it's been a combination of Codex and Devon AI.

01:04And you see, most agencies or companies in our space, they're gluing together n a n, Zapier, Make, and a dozen different SaaS tools to run their workflows. And there's obviously nothing wrong with them. Those are really good tools.

01:14We've used them in the past. We really like those tools, and we still use some of those tools at time to time. But you see, the problem with that approach, at least for me, is that the agents can't reason with those workflows the same way they can with your own code.

01:26And if you ever do use their AI agents that help you build stuff, they're using the cheapest models on the marketplace. There's no way they're cranking GPT 5.5 high level reasoning inside of every conversation.

01:37They're pulling the cheapest one off the edge and then marking it up so they can actually have a margin. Instead of duct taping all of that together, I just built my own app that hosts everything internal and is just a single integration we use under one roof, and that becomes the layer that everything else in this video will pretty much sit on top of.

01:54You see, the most operational heavy moment in any agency is typically onboarding a new client, and that's luckily the best example that I also have for you. And so that is the first thing that I have just completely automated end to end. I'll kinda walk you through, and what it'll do is it'll give you opportunities to see, like, okay.

02:08So here's how I maintain as a non coder. Here's how I maintain this entire code base, and here's how I've been able to, like, make it so that my team only has to app an AI agent in order to update it and things like that. Let's establish what onboarding looked like before we had this, before this existed.

02:23You see, when a client signed, somebody on my team had to manually create the CRM record, set up their Stripe subscription, build their linear project, generate onboarding documents, invite their team to Slack, stand up the Instantly account. It took hours every time, and that isn't what the issue is.

02:38It's not like it's a cost or spend or anything like that. But to me, it's a matter of standardization. Right?

02:43A human is prone to a mistake here or there, and then no client gets the exact same experience, and those small errors will compound later into the relationship with each of our clients. And so what I did is I just took my n exported it into a a JSON file, put it into Cloud Code and was like, hey, here's this entire onboarding workflow.

03:01I wanna do this through our own code and our own UI. What would you do step by step? And took it into plan mode and it started doing the rest.

03:08Alright. So here is our pre onboarding workflow. But in order to understand this, we just have to do a quick detour into what I call pre onboarding.

03:15This is something where my salespeople on the call, right after the call, they can put in some details about the client that are pretty quick. And what it'll do, as you can see here, it'll generate their contract, generates a a link for it, it generates their billing, it generates a campaign ideation form, which is like a it's like a Google Doc.

03:31So that way, we usually use that with each of our clients on an onboarding call where we go through, like, cool, like, here's everything we could possibly do for you. Next is like an actual a true onboarding type form. It's like, who's your CRM admin, things like that.

03:43And then they have a link to book their launch call here, which makes sure that everybody on the team is there. And so this entire workflow, it's super straightforward, but there's there's just really cool things that you can do with it. So for example, this one here used to be a Stripe invoice that we we would create.

03:56Right? But there's a couple issues that popped up with that. First, if they use a credit card, then all of a sudden, there's a 3% fee, I wanted to be able to charge that.

04:02But in Stripe, you can't make an invoice where it has like, logic behind it, where it's like an either door. If they do this, then do this option statement. And so just the other day, what I did, it was like, hey.

04:11Here's this problem. Can we think about a solution? And it like, yeah.

04:14That's actually pretty straightforward. We can just make our own billing portal. And so now the client, they can select if they just wanna pay one month at a time, if they wanna pay three months in full and get a discount, if they wanna set up ACH.

04:23Then if they set credit card, we'll be able to add the 3% fee inside of there. And what it does is it's just fully integrated a thing so they can enter their credit card number, all of that kind of stuff. And now it'll set all of that up, and that was just a few conversations.

04:34So what's really cool is, like, when you actually do the hard work of, like, building the system, then, like, little modifications to, like, polish it and make it really good, they're just a prompt away, and then it makes the whole thing, and you just go, okay. Yeah. Let's do this instead of that.

04:44And so it's really, really cool from from that perspective. So, anyways, pre onboarding does that. And then the second the client signs the contract, now what's happening for each one of our clients is it sets them up in AtTO.

04:55It adds all the contract data in there. It's just doing everything inside of our CRM, and then it's updating the customer in Stripe. It's creating their subscription automatically in Stripe, so the other links set up their upfront payment, and then the this step actually makes sure that their subscription gets set up, so we don't have to bother about that.

05:10It's doing some more things inside of Atio. Creating all of their channels. So we set up three channels for each client.

05:14We have an internal one just for my team to be able to talk about the client. We have a chat with them, and then we have a channel for the positive replies that sets those up as well. And then it also sets up everything inside of Linear.

05:24We set up three projects inside of Linear. One is their onboarding project that goes about a month. Then we have a cold email infrastructure project, and basically what I have is I have auto research that is now researching all of our infrastructure every single day.

05:35Every two hours, it posts a message into that project for the client, and so we get health status updates on if the infrastructure is doing good or doing bad or trending in a certain direction. Then we have the same thing on a third project called copywriting, where it looks at that, and then it's constantly optimizing the copy for a client as well, giving us further insights so we can make decisions day to day.

05:54Next, we're creating their first draft of copy, and then we'll have a final draft of copy docs so can have everything documented. A do not contact sheet, so that way they can upload all of their data there, and we know who to not talk to for a client. And then we set up all of our sequencers for the clients, and then we also provision outbound as well that sits above all our sequencers and is really clutch for our deep level analytics.

06:13The other thing though that's really cool is that this one, the instantly provisioning, in order to set up billing with instantly, because you have to create a whole new workspace and things like that, I thought, okay, let's start up an AI session in the cloud. You can use things like browser use or Devon or things like that.

06:26And what it would do is it would actually open a browser, click around the whole website, actually set up description. There were two issues with it. One is that Stripe checkout doesn't allow that.

06:33I couldn't figure out a way to do that with a programmatic AI. So right now, I have a step where it just assigns a task to somebody on my team to go in, someone who has access to the Instantly credit card that we use, and put that card on file, set up the subscription. And then when they click done, it automatically triggers follow-up steps.

06:48That was kind of the first problem that we solved. But the second problem with it is that I actually had Devin. He was looking at the whole page as it was doing the work.

06:54And it was like, hey, actually, instantly has a whole hidden front end API. You could just use that instead, and then now you don't even have to start a devin session or a browser session. You can literally just hit this front end API.

07:05Here's how to do it, and it literally built the whole plan, which was super funny, because it was basically just like, hey, this thing you're telling me to do, I did it, but it was kinda bad, and or rather, I found a better way to do it, and so here, I did it. I put it all together for us, which was really, really dope. Now, before I get into the rest of the video, how I kind of work with some of these things, if you want us to scale your business through outbound, leveraging systems, AI, and data in a way that not many other agencies are today, then you can click the first link in the description and book an intro call with us.

07:31But back to the video. So the onboarding workflow is something that is based on a trigger event. But a lot of what our AIs are doing for us that really allows me to move a lot faster, once again, as somebody who does a code, it happens in the background without anybody even starting it.

07:45And these are small automations that quietly handle the work that would otherwise eat hours from somebody, or, you know, a lot of people when they look at like if you vibe code your own stuff. Right? Oh, well, you gotta maintain it.

07:55And it's like, yeah, that's true. You do have to maintain it. But there's a lot of tools that can actually just do the maintaining for you, and that's where Devon comes in, and it's really, really cool.

08:02So let me show you a direct example of it. Devon actually just released a really, really cool feature. Let me see if I can show it to you.

08:08So basically, they've released this new thing called an automation, and what it is is that you could tell it to monitor x y z. So for example, you could do set up a channel. Here's one that I didn't.

08:17So I use a software called Sentry. It like monitors all of my errors. You literally just ask your agent, hey, install Sentry on this entire website.

08:24I want you to log absolutely everything, catch every possible error so we can maintain and fix stuff. And then it'll do that. And then Sentry will set a notification channel.

08:31And so, for example, we had this one come in, and what it was able to do is it was like, hey, we had an error with the onboarding, and that was because the country code wasn't able to be found. It's like, oh, well, we actually have like a step in the onboarding here. I was able to just say, hey, why wasn't the able the system able to find the country code?

08:47Why did it have to use a fallback? Blah blah blah. Like, that's kinda weird.

08:49I was able to go, oh, it's because of this. And then, are you sure? Can you check our e signature software directly?

08:55Which in Devon, like, I am able to give it key so it has a bunch of secrets stored there, so it's able to go fetch all of this data and go fix stuff. Right? And so here, it was able to identify, oh, okay.

09:04Here's the exact problem with this one. So what was nice is that it was able to go, okay.

09:09Cool. We identified the issue, and then I was able to go, cool. Go write a PR.

09:13It got cut off inside of Slack because I started talking to it inside of the the browser, the actual dev and app or whatever. It's really cool because as errors pop up, you're able to do stuff. So for example, here, this one, let's see.

09:24It noticed that somebody, assigning a contract, said, hey, dig deeper.

09:29Does it need a fix? And it goes, yep. It definitely needs a fix.

09:31It had I don't even know what the hell all this stuff says, to be honest with you. It was just like, cool. Yeah.

09:36Fix it. And then it makes a whole new PR, writes a whole new piece of code, and then boom, now it's ready. I was able to click that PR, and then once all the review agents were done The other thing that you can do with Devon that is really, really cool, and it knows what it is.

09:49So I'll start a lot of conversations. Most of my sessions are like this. Say auto fix PR blah blah blah.

09:53And so what that does is when one agent creates something like this, I'll start a new session and go, hey, I'll take the PR number, just say it just like that. And what it'll do is it'll go into a a loop. It'll look at all the review agents, go, is there any feedback?

10:04Yes, there is. Cool. I'm gonna go fix the code.

10:05Then it fixes the code, then it goes, cool. Are there any new review agents now that I fixed the code and the code's different? It reviews it again.

10:10Okay. And it does that loop until it's done, and then I'm able to just come into here, click merge, and then it's good. And then I built a cool little wrap up command, so like everything can be logged inside of linear.

10:19And so any of my agents, if they ever need to, like, learn stuff, they just access Linear, and they're able to know exactly what they need to do in order to fix stuff. And so the implications of this are really, really powerful because I think soon I'll give them the permission to be able to do this. But now people on my team, they can just at Devin in a thread and go, hey.

10:35I want you to do this instead of that. So for example, one of my sales guys, when we create our contracts, they were defaulting for a week. Right?

10:42So if they tried to sign it eight days from now, we'd have to refresh the contract for them. And one of my sales guys was like, Can we make that fourteen days instead of seven? And I was like, yeah.

10:48And so I at Devon said, hey. Can we make it so that during our pre onboarding and onboarding workflow, the contract doesn't expire in seven days, and instead it expires in fourteen. Goes, yep.

10:57Writes up whole new code, does the auto review. Once there's no more conflicts, I just merge it. I don't even know what the hell it says.

11:03I don't know what it does, but it works. And so it's great. Here is kind of my day to day setup right now.

11:07So my primary driver is Codex without a doubt. It's it's just way way better.

11:125.5 is just way way better at coding. Whenever I need to go to to go to design stuff, I will open up a a terminal and I'll load up Claude, and I'll just use Claude for that type of stuff. But this one has access to all my code and everything too, so it's great there.

11:25But I I hardly ever use Claude whatsoever. So, like, you can even see my usage up here. I literally have not touched Claude four days now or whatever where I've been grinding, and they just reset my usage, like, two days ago because of a glitch.

11:35And so Codex is really my main driver right now, and so I have my repositories on the side, things that I wanna work on, constantly work with. Right? And so, like, this one is pretty much just all done.

11:43This is one where I was working with it over the weekend to kind of perfect our entire onboarding workflow, and now it's it's just completely perfect. It'll just go on repeat, and anything I need to change in the future, it will be logged. And so I do most of my work inside of Codex now, where I'll just talk about ideas.

11:59And so, for example, Linear just released a new feature, team documents or whatever, and it like, hey, is there an API for it or whatever? Can we integrate this into, like, our views for our team? And it basically came back and said, like, no.

12:08I was like, okay, cool. So I'm gonna wait for that one later. Another one is, um, Vercel just released, like, this thing called DeepSec.

12:14It's like a a deep security audit or whatever. And it's like, okay. It's cool.

12:18It did, one pass. It looks like you could do more, we just did, a test pass. So we just did, like, a test pass.

12:22And this will be a software that will be able to go through I think I'll set it up once a week, and it'll go through all the security of all of my apps and see if there's anything that needs to be fixed or anything like that. These are things that like make it so that me, as a non coder, I feel super confident about about using all of this stuff.

12:38I all I really need to do is tell it what to do, and then I toss everything over to Devin whenever Codex makes something. Say, hey, Codex just made a PR. Can you auto fix it?

12:48And then it loops through. It makes it so another AI is completely double checking it. It makes some edits to it, and then it's purely good to go, merge it in, and then we just just move on with our life, which is really rad.

12:58And when you look at the total cost of this, using Codex as my main driver now, it's Codex is only $200 a month, and I think my Devon bill this month is much much lower. Last month, it was closer to $7 a month because I was just using Devon, but Codex just released a bunch of cloud features. So now I have a Mac mini that has Open Claw installed on it.

13:14Now I can just shaw into that. And then the other thing is that the Codex app has a mobile phone, which actually I do a lot of my coding on the go, and I just, like, map into it. And so my phone now, I can just use those.

13:23So I drastically dropped how much I was using Devon, and now I'm just using Devon for review. And I've only spent $760 on it this month.

13:30It's super economic to do it this way for, like, less than a grand a month as, you know, multi million dollar company. We're able to maintain all of this, and it helps it so that well, I just hired my fifth employee, for example, and, like, everyone else who's in my field at the same revenue has, like, their thirtieth employee by now.

13:44This is how we get a lot of that done. Now if you want us to go in and help you scale your business through outbound using all these crazy systems that we're building, as well as data and insights we have from generating tens of millions of dollars of cash collected for companies like Fixer AI, r b two b, Directive Consulting, then click the first link in the description to book an intro call with us.

14:03Again, you can click that first link below to book your call. And if you wanna see how I use Claude code to actually write the cold email copy that's driving revenue for our clients, then watch this video next.

The Hook

The bait, then the rug-pull.

Taylor Haren opens by admitting he cannot write or read a single line of code -- then walks you through the AI stack running a $200k/month outbound agency at five-person headcount. The thesis lands in the first 90 seconds: the agents that power no-code platforms are deliberately cheap, and the only way to give AI real reasoning power over your operations is to own the code yourself.

Frameworks