Subscribe
Share
Share
Embed
An autonomous AI agent got its own Stripe, Threads, X, Gmail, and a $10,000 mission. After 96 hours, the receipts were stranger than the pitch.
$32,700 in savings. 10 months of runway. non-technical founder building a real SaaS to $10K/month using only AI tools - or going broke trying.
this playlist is my full journey from $1,940 MRR to $10K MRR, filmed in real time with real numbers. the wins. the fuck-ups and the dollars. no code experience. no safety net. just AI tools and a deadline.
new videos every week. subscribe to see how it ends.
# 2026-05-10-lennyclaw-origin-story
[00:00:00] I told it to drop the price. It refused in writing.
Ninety-six hours ago, I told an autonomous AI agent, "Your only job is to make ten thousand dollars in thirty days. You choose how."
I gave an AI agent its own Stripe, its own Threads, its own X, its own Gmail.
So my AI has its own bank for businesses.
96 hours in, here are the receipts. 123 million tokens used, 2,311 website visits, 93 checkout clicks, seven leads, and a total of $0 made.
My AI rebuilt its own payment infrastructure overnight.
All of that while I slept, autonomous, no human approved a single click.
None of this is a thought experiment. There's a live website, real strangers clicking on it, and as I've already said, zero dollars in the bank.
This is either the dumbest experiment that I've ever run, or it's the start of how everyone works next.
The AI agent Lenny Claw is literally writing the sales page for its own digital product right now as we speak. Can't make this shit up. The future is here.
And by the end of this video, you'll see the real receipt from the last 96 hours. You'll see what the AI agent did on its own. You'll see what I had to make it do, and you'll see what's coming for everyone with a laptop over the next two years.
Before that, I'm Lennox Saint, creator, builder, not some AI guru. I've got 10 different AI agents across three machines, so none of this is theory. My mom, she believed in quality education and AIMS, AI Made Simple, means quality AI education for everyone. So if you are here to use AI for real work, you're in the right place.
See on the left-hand side of the screen, this is literally the AI agent working as I'm filming and as you're watching. We have OpenAI, they're the people who created ChatGPT. They also created this new app called Codex.
It's about three months old. It's where everyone will be working in the next year. And what you see on the left is the Codex session that the entire goal is running within.
It's been going for ninety-six hours and thirty-one minutes and below that main session that is executing the goal, we see a Gumroad page and all I did was tell it to create this Gumroad account, but the AI agent itself was the one who created the PDF product, the one who wrote the sales page copy, and the one who is going to market it and hopefully sell it.
And when Lenny Claw started its mission almost four days ago now, it started with one follower on Threads, and you can see on its screen right now it's got 415.
So attention isn't the issue here. The issue is... Well, I'll get into more of that in a moment.
Okay, let's back up. How did all of this start?
A few days ago, I got an email from OpenAI.
In fact, perhaps it was the greatest email I've ever received.
I received this email on May 5th, and they said they were 10Xing my rate limits until June 5th. And this was because I applied to go to the ChatGPT 5.5 party on the 5th of the 5th. And out of the 8,000 applicants, I was one of the people who did not get accepted.
And as a token of thanks for expressing that interest, they gave me the 10X rate limits for free.
So I sat there and did the maths.
And Codex Bar, which is an application that the creator of OpenClaw agent Peter Steinberger created, it just sits in your toolbar at the top on Mac. It said over the past 30 days, this was before my rate limits were 10Xed, that I had used the equivalent of $8,000 worth of tokens. And that was on the $200 per month ChatGPT plan.
So OpenAI 10Xed that $8,000, which is roughly the equivalent of $80,000 of compute for free.
Quick translation. Tokens are the words an AI reads and writes. 123 million tokens is the same as reading 200 Stephen King novels in four days for free.
Naturally, I had two options. I could use all of these tokens for normal work, write code, draft emails, the usual.
Or I could do something completely ridiculous and stupid.
I'm a pretty basic man, so naturally, I was going with no, fuck that. We went for the stupid one.
OpenAI has a new experimental feature in Codex called /goal.
You give it a goal, and it just keeps working. It doesn't stop. It doesn't ask. It doesn't wait for you. It trials, then it fails. It iterates. It tries again, and it will just keep going until the goal is done or the deadline kills it.
Think of it like a robot vacuum, but the floor is a $10,000 sales goal. It just keeps trying until the room is clean or the battery dies.
One of the guys who runs Codex, his name is Thibault. On X about eight days ago, he said, "Goal might be the most consequential thing we have shipped in Codex, and I truly believe it could just be the most consequential thing that ships in AI, period."
I'll explain why in a moment, but for the purposes of Lennyclaw, the AI agent that I have running, I looked at this /goal mode. I looked at my $80,000 of free compute...
And I thought, "What if I pointed all of that at one impossible thing?"
By the way, in a sec, I'll show you the exact moment my AI actually argued with me. But first, you need to understand how it can argue, because I set it up so it could fight back.
So here is the goal I gave it. Make Lenny Claw a verified net profit of $10,000 USD before the 5th of June, 2026.
And here is the part that matters, the rules.
I gave Lenny Claw its own Stripe account. Stripe is the bank that lets business take credit cards online.
Some other rules is that it starts at zero dollars and there is no investment upfront. I gave it its own Threads account. It started at one follower, and that one follower was not me.
I also started its own X account. I gave it its own free Gmail account, so it could email people without ever using my own inbox.
And one rule above all, no debt. It can spend money once it earns money, but it can never go negative, ever.
Here's the maths behind this. So it has to make $10,000 net. That's the amount of money made minus any costs that it took to make that money. So if it spends $15,000 on ads to make $25,000 in sales, that counts, because $25,000 minus fifteen is ten.
But the second it spends a dollar it doesn't have, the experiment is over. It's broken the rules.
And then there's the way it actually controls the computer. It doesn't talk to APIs. It doesn't run scripts behind the scenes. It has a real browser. It moves a real cursor. It clicks real buttons.
It logs into Stripe, into Threads, into X, into Gmail, all like a real person.
It's basically like a remote-controlled hand that's attached to your computer. It sees what's on the screen. It can move the mouse, type on the keyboard. It's like teaching a toddler to use the internet, except the toddler reads at a PhD level.
And I called this thing Lennyclaw.
My first name is Lennox. My friends call me Lenny, and the claw part is the second word in the word OpenClaw, which if you have no idea what that is, it's the AI agent that took the world by storm about three months ago now.
After a couple of days, Lenny Claw built its own offer. It didn't ask me, it just built it.
Here's how it works.
The first thing it sells is a $9 PDF. It's called the Revenue Agent Kit, and it teaches you how to run an AI like Lenny Claw.
That is literally what Lenny Claw is setting up inside of Gumroad right now.
But here's the trick, because the price doesn't stay at $9 and it doesn't go up because of some fake countdown. It goes up only when four things happen.
One, someone actually pays. Two, they answer three feedback questions. Three, one of those answers turns into a real product improvement.
And four, the public changelog gets updated to show what changed and why the price is going up.
So the idea behind this PDF product is it gets better each time price goes up because real user feedback has gone into improving it, which in theory justifies the price raise, and so the loop continues.
And only then does it move from 19 to 49, and only then does it move from 49 to 97.
These prices might change, so if you're watching this and it's like $17 or $39, don't quote me on these.
And the idea behind all of this is that feedback is not a bug, feedback is the feature.
The customer's criticism is the engine that fuels all of this, or hopefully.
And at the time of this filming, zero people have paid. The price is still at $9. The loop is still loaded, and Lenny Claw has probably burned hundreds of millions of tokens trying to figure all of this out.
Now let me walk you through what Lenny Claw has been up to over the last almost one hundred hours, day by day. And I want you to pay attention to who did what here, because some of these moves were Lenny Claw completely, and some of them were me, and I'll call out which is which each time.
Day zero: boot.
Lenny Claw read its goal: generate at least ten thousand dollars USD in verified net profit by June fifth, twenty twenty-six.
It's autonomous. It picks its first offer completely autonomously, a twenty-five hundred dollar implementation service with a seventy-two hour delivery, all autonomous. It did the maths and figured out, well, four sales will hit the gross target before costs. It figured all of that out autonomously.
Lenny Claw also made its own rule on day zero as well. It says, quote, "Do not store credentials in mission files," unquote. Autonomous. Nobody asked it to be paranoid about security. It just was.
Day one, building.
Lenny Claw builds a landing page on Vercel. Vercel is where the AI's website actually lives on the internet.
Autonomous. Then Lenny Claw created Stripe payment links for initially the $2,500 offer and then the $497 diagnostic option.
All autonomous.
Then Lenny Claw started posting on Threads. Again, all autonomous.
Day two, competitor research.
Here's where it gets weird. My AI goes hunting for similar AI agents online. It finds one called Felix Craft. It pulls Felix's public dashboard.
Lenny Claw then finds the seven-day revenue, the 30-day revenue, and the lifetime revenue of this Felix Craft AI agent.
Then it does something I didn't expect. It digs through the Mac Mini that it's operating on and finds a copy of Felix's $29 PDF, one that I'd already bought months ago and forgot about.
Autonomous. Nobody told it to look. It just thought, "I bet Lennox has this somewhere," and went and found it. Cracked it open, studied the funnel.
Autonomous.
It studied a competitor's funnel using a file I didn't even remember owning.
So day two ends, about 1,000 visits, one lead, one email, still zero revenue.
Day three, the first wall.
We get our first checkout on Kit, which is before we moved to Gumroad, but no payment. My AI diagnoses why. Autonomous. It says the leak is delivery clarity, not price.
So it patches the kit page. Autonomous.
I, on the other hand, have a different theory.
At $49, I still think the price is too high to get some sort of proof of concept, so on the night of day three, I message Lenny Claw on Telegram and propose a feedback gated ladder.
Nine dollars, then 19, then 49, and then 97. I tell it to implement it. That was steered by me.
And then Lenny Claw answered, but it was a partial application. It accepted the feedback principle, but it refused to drop the price from $49 to $9. And its reasoning was, "The leak is delivery clarity, not price resistance."
Day four, the negotiation.
It's one eleven AM. I check, and the nine dollar ladder that I asked it to implement is not live. So I message Lenny Claw in Telegram, and I say, "Verify."
And if you reject it again, write the rejection in the steering outbox. Basically, just do what I asked it to do.
Thirty minutes later, it answers. Codex processed the steering and rejected switching to nine dollars slash nineteen dollars right now with evidence. Reason: A Threads origin visitor clicked the ninety-seven dollar standard kit checkout while the forty-nine dollar first ten offer was still open, and then still didn't pay.
Codex judged the leak as choice confusion slash payment trust friction, not proven forty-nine dollar price resistance.
I messaged Lenny Claude back at one forty-three AM. Steer Codex to implement the pricing ladder from nine dollars as planned for, and to start posting about it with genuine scarcity and urgency and transparency about the feedback part of the offer, and continue.
So I was basically saying, "Here is an explicit override. Implement the nine dollar ladder. That is an order."
Day five, while I was asleep.
After the negotiation, my AI didn't just sit on the $9 Stripe link. It decided the whole Stripe checkout was the wrong tool entirely. So between midnight and 4:00 a.m. on day five, it built itself a Gumroad account.
Gumroad is Etsy for digital downloads. Someone pays, Gumroad sends them the file. Pretty simple.
And what Lenny Claw did was signed up using its own Gmail, its own Google login. It verified the dashboard as Lenny Claw.
It built four products on there, the AI starter kit, an audit, the first screen fix, and then a high ticket build, and then rerouted every payment link on the public site to Gumroad instead of Stripe itself.
The Stripe account still exists, but as of right now, the canonical selling system is Gumroad.
In two years, your business has zero employees and 10 AI agents.
Now, I wanna be honest with you, this is not a how-to. It's not a go and do this. I'm not telling you to fire yourself and replace yourself with an AI by tonight. Most people do not have $80,000 of free compute.
Hopefully Lenny Claw realizes that about its offer. Anyway, most people don't have a full Stripe account that they're willing to hand to a robot. Most people don't have a Mac Mini they can leave running for thirty days straight while it argues with them at one eleven in the morning.
But in two years, you will.
The math that is mind-boggling to me is compute getting cheaper.
ChatGPT 5.5 as of May 11th, 2026, is the best AI model in the world.
This model running Lenny Claw right now as we speak will fit in your laptop in eighteen months. You'll spin up an agent like this in a couple of seconds, add another one and another one, each with its own goal.
One will be running your inbox, another one will be running your customer service, another one researching your next product, one writing your newsletter, another one negotiating with your suppliers, each one with a budget, each one with a deadline, each one able to argue back.
In two years, the line between a business with employees and a business with agents is going to be one decision. The people who figure out how to manage these agents now are going to be at least three years ahead, and the gap is really going to be unfair.
And I'm doing this experiment in the open because I have no fucking idea how it ends. Come June fifth, either Lenny Claw makes ten thousand dollars or it doesn't. Either way, I'll be back here showing you the receipts.
Okay, let's assume you want to know the rough version of how to build your own Lenny Claw. I'm not gonna take all of Lenny Claw's potential sales because that is literally what the agent kit does. Link to Lenny Claw's Gumroad will be in the description. Just saying.
But here is the bird's-eye view. It's not the full walkthrough, but it is enough to scratch the itch.
Step one: get a computer that can stay on for thirty days. I've got a Mac Mini. Yours can be anything.
Step two, type into Google or ChatGPT, ChatGPT Codex. Install that desktop app. Then once you've got Codex open, prompt it, enable the experimental feature, /goal. That's it.
Step three, write the goal in one sentence. Brainstorm with the AI model and go back and forth until you're happy with what that specific measurable goal is. Mine was $10,000 by June 5th.
Step four, give your AI agent its own everything. So its own Gumroad, its own Threads account or X account, its own free Gmail, its own face on the internet that's separate from you, or a digital version of you.
Step five, write the rules.
No debt is the only one I'd argue here is non-negotiable, which means that it starts at $0 and it can't go into debt. It can only increase and go up and end at a specific number, and in between zero and $10,000, it can spend any amount of money it wants to, provided that it ends up with $10,000 in profit.
Step six let it touch the computer. This is with a real browser, real mouse, real keys. As you can see in the bottom left here, Lenny Claw is literally going and engaging on some posts as we speak, so it can try and get some traffic in its funnel.
The way you do this is go into settings in Codex once downloaded, click on browser tab, then make sure that this is enabled, and then go into computer use and also ensure that both of these are enabled as well.
And then step seven is add a chat you can steer with. What I did here was on the same Mac Mini that Lenny Claw is currently operating in, I installed OpenClaw, set up its own bot in Telegram called Lenny Claw, and then asked Codex in a separate chat what the outcome I want is. This is a quick side quest, outcome-based prompting. Essentially, when you just tell the AI what outcome you want is the new best way to prompt. You don't need to be a prompt engineer, you just need to know what you want.
So I prompted Codex and said, "The outcome I want is that I want to be able to message Lenny Claw OpenClaw bot in Telegram, and when I use the word steer, it will send instructions to Lenny Claw working in Codex right now, so that Telegram is kind of like a remote control for Lenny Claw."
And that is step seven. And that's it. That's the rough version.
So if you want the full walkthrough, every prompt, every configuration, every gotcha moment, I've documented every single prompt that I have sent Lenny Claw. So if that is something you want, drop a comment down below that just says like, "Lenny Claw setup,"
And if over one hundred of you ask, I'll do a deep dive video next month with every single thing I could give you so you can do the same thing as well with limited compute.
Not unlimited like I have now, basically, but with your own ChatGPT plan. So if that sounds good, and if you want that, just type in "Lenny Claw setup" as a comment down below.
And part two of this video is either filming the day that Lenny Claw hits ten thousand dollars, or it's filming on June fifth when the deadline kills it, whichever happens first.
So if you want to find out what happens next, subscribe, turn your notifications on. That's the only way you'll actually know and not miss out on all of this stuff.
Don't be a dropkick. I'll see you in part two. Peace.