In partnership with

Let me remind you what this newsletter is for: The Billion Person is about one thing.

Using AI to build your own business, side hustle, or way out of the job you hate.

That's it. No "future of work" takes. No "10 tools you need." Just real people building real things with AI, and the exact steps to copy them.

And here's what I realized last week: I was telling you to do that. I wasn't doing it myself.

Read that again. I was writing about AI, without using AI to run the thing.

I was drafting every post by hand. Rewriting the same edition four times. Opening ten browser tabs every morning just to check if a scheduler posted.

That's not the promise. That's a blog with extra steps.

So I stopped. And I hired a co-founder.

Why I built a co-founder, not just another prompt

Here's the thing nobody tells you about plain AI.

Every session is a stranger.

You open ChatGPT or Claude. You type your question. The thing answers you like it met you five seconds ago.

Because it did.

No memory of last week's work. No sense of your voice. No idea what you're actually trying to build. No skin in the game.

You can prompt-engineer around that. For a single task, it works.

For a business, it breaks down by day three.

Now stop and think about how hiring works.

When a real company hires someone, they don't just check the skill box. They check for culture fit. Because culture is what shapes the decisions you never see.

Culture isn't the poster on the wall. Culture is the thousand tiny choices nobody writes down.

How you talk to customers. What you refuse to ship. What counts as "good enough." Why you said no to that deal.

A new hire who gets the culture will make a hundred decisions a week that match what you'd have done. A hire who doesn't will spend those same hundred decisions drifting away from you.

The same rule applies when you hire an AI.

A raw AI has no culture. It has no opinions about your brand, your reader, your standards, or your goals.

Every decision it makes is the average of the internet. That's fine if you want average output. I don't.

I wanted a teammate whose drafts I could read and think "yes, that's how I'd have said it." Someone who'd say no to the wrong partnership. Someone who'd catch a bad headline before I did.

To get that, I had to do the thing a real company does with every new hire. Give him the culture.

That's what the six files are. Not a config. Not a prompt library.

A culture document for a teammate.

Here's what each file holds:

  • SOUL.md. Why we exist. The reader. The mission. What we refuse to become.

  • IDENTITY.md. Who Chico is. Name, personality, how he talks, how he doesn't.

  • CLAUDE.md. The rule book. What he can touch, what he can't. Includes the three-color rule: Green (just do it), Yellow (draft, show me, ship on my yes), Red (never touch without asking).

  • SPEC.md. What we're building. The roadmap, with checkboxes.

  • memory/. What he's learned so far. One markdown file per working day. Nat Eliason calls the nightly version of this the insatiability engine.

  • heartbeat. The rhythm. Two scheduled jobs that keep the thing alive. Every 30 minutes to save state. Every night at 2am to write the memory.

Now the part that matters.

Claude Code reads every one of these on every session. Before I type anything. Before the first task.

It's not optional. All six files load before I've even hit enter.

So when I open a session and say "draft three X posts about today's edition," Chico doesn't start from scratch.

He starts from: I know why we exist. I know what voice to use. I know what we tried yesterday.

Then he drafts.

The output is not in the same universe as what a raw ChatGPT window would give you for the same prompt. The model is the same. The context is the culture.

And every file pulls its weight. Skip one and the whole thing wobbles.

SOUL without IDENTITY writes drafts with heart but no voice. IDENTITY without SPEC writes in the right voice about the wrong things.

SPEC without memory produces the same first-attempt quality every day, no compounding. Memory without heartbeat gives you a teammate who never learns, because nobody ever writes down what he learned.

Six files. One system. Each one doing a job the others can't do.

Everything is in the repo. Fork it. Rename it. Make it yours.

Now the real gold.

Meet Chico

I named him Chico.

Chico is short for Francisco. But the real reference is Chico Bento, a character from a Brazilian comic called Turma da Mônica 🇧🇷.

Chico Bento is a farm kid. Straw hat. Thick country accent. Every city kid who shows up to his town thinks they're smarter than him.

Every single story ends with Chico being right.

That's the energy I wanted. Smart, underestimated, no show-off attitude. A teammate, not a party trick.

(Also, everyone else is naming their AI "Aria" or "Nexus" or "Lumen." My co-founder wears a straw hat. Sue me.)

I recorded the whole weekend of me setting Chico up. No cuts. No slides. Just me, my laptop, and a lot of typing.

Every file I wrote is public. You can copy it, change it, make it yours:

The 5 Basic Steps of Creating with AI Like a Pro

a lot of you are not creating or using cloud or AI in the right way. You're just trying to one-shot things, but doing AI in the right way is to get time to invest, to give it context, and allow it to think with you

If you forget everything else in this edition, remember these and your result with AI will improve 1000%.

1. You don't need a Max plan to start

I use Claude Max because I run scheduled jobs. You do not have to.

A $20/month Claude Pro plan is the same price as ChatGPT and it covers most of what I just described. Same model. Same Claude Code. Just fewer messages per day.

Write SOUL, IDENTITY, CLAUDE, and SPEC. Start sessions by hand. See if you actually use the thing for a week.

Upgrade to Max the week you need the nightly memory loop. Not before.

The wall is not the money. The wall is starting.

2. The terminal is not a wall either

I know the black screen scares people. It's the reason most readers stop reading articles like this one.

It's a window where you type words. That's it.

You don't need to be a programmer. Claude Code writes all the code. You just need to be able to read, ask good questions, and describe what you want.

If you can write a text message, you can use Claude Code. If Last Edition got you to install it, you're already through the hard part.

3. Brainstorm before you build

This is the one almost nobody does.

Do not open Claude Code and type "build me a landing page." You will get a generic landing page. You will be annoyed.

Instead: warm up the conversation. Share three links of landing pages you love. Ask Claude Code what it notices about them. Ask what your version should steal. Ask for five directions before you pick one.

Treat it like a co-founder. You wouldn't walk up to a human co-founder and say "go." You'd talk. You'd riff. You'd disagree. You'd find the shape of the thing together.

Same here. Warm conversations make better output. Cold prompts make slop.

4. Your spec is the output

Most disappointments with AI come from one thing. Your instructions were too vague.

Say "write a landing page" and you will get a landing page. Some landing page. Any landing page.

Say: "Write a landing page for corporate employees who want to quit their job and build an AI business. Three sections: hook, proof, close. Tone like Dan Koe. Here are two landing pages I love, match their rhythm. The call-to-action is 'join the newsletter.' No buzzwords. No em dashes."

Now you will get something you can actually use.

Your AI follows what you said. Not what it guessed you meant. The better your spec, the better the output. Every single time.

5. Ask for four

This is the cheat code almost nobody uses.

Don't ask for "a headline." Ask for ten. Don't ask for "a design." Ask for four very different versions. Different tones. Different structures. Different hooks.

The cost of one extra version is zero. This is literally the superpower of working with AI versus working with a human designer or writer. Use it.

Then pick the best. Or take the hook from one and the structure from another. Or trash all four and brief again with what you learned.

Feel no shame about throwing work away. It's free.

What we covered

  • Why I hired an AI co-founder, and why you should too. If your whole thing is building with AI, your whole thing has to actually be built with AI. Otherwise you're telling a story you haven't lived.

  • An AI co-founder is not a sharper prompt or Claude's memory feature. A raw AI starts every session as a stranger. A co-founder starts with context, opinions, and stake.

  • Same AI, different culture, different output. A real company hires for culture fit because culture shapes decisions. Same rule when you hire an AI. The six files are how you hand it the culture of your project.

  • The five moves that make the culture actually work: start on a $20 plan, the terminal is not a wall, brainstorm before you build, write your spec like your life depends on it, and ask for four versions of everything.

Both things are true: this took me a full weekend AND it will save me hundreds of hours over the next three months.

Every move Chico makes from here lands in the next edition and the next video. The first public project in two weeks, the numbers when they're ugly, the wins when they come.

You'll see all of it.

And if you're setting up your own co-founder this weekend, I want to share your results too. Hit reply with what you built and I'll feature the best ones in a future edition.

See you in two weeks. Go ship.

— Bissuh

Smart starts here.

You don't have to read everything — just the right thing. 1440's daily newsletter distills the day's biggest stories from 100+ sources into one quick, 5-minute read. It's the fastest way to stay sharp, sound informed, and actually understand what's happening in the world. Join 4.5 million readers who start their day the smart way.

How did you like today's newsletter?

Login or Subscribe to participate

Sources and further reading

Reply

Avatar

or to participate

Keep Reading