The New Poverty Line Is $100

AI and the future of earning power

On June 19, 2009, K-pop girl group Girls’ Generation released a teaser for Genie, their upcoming second mini-album, and your girl really wanted a copy.

I wanted to play the CD in the car on the way to school. I wanted to blast I’M GENIE FOR YOUR WISH in my room. Most of all, I wanted it for the photo cards.

K-pop albums had this tiny blind-box mechanic built in long before Labubus became a household thing: buy the album, peel open the packaging, and find out which member’s face (from which scene!) had been tucked inside. If I were lucky, I would get Tiffany in the short razor bob and the white marine uniform.

But K-pop was not big in Indonesia in 2009, at least not in the walk-into-a-store-and-buy-it way. The only way to get the album was online, where the price was inflated because it was imported and niche and not treated like a normal Indonesian CD. There was also one small problem. I did not have a credit card. Or the money.

I was merely thirteen.

I could have asked my parents, but it did not feel like a reasonable thing to bring up. They would have stacked it against everything else: books, tuition, new shoes. If you were my dad, you would have stacked it against buying stocks. (He had me read Bisnis Indonesia every day to check prices.) A Korean girl group CD with a collectible photo felt, in that household economy, kinda silly.

So I did not buy it.

This is not a sob story. Nobody needs to feel sorry for a teenager who missed out on a photo of her bae. But I remember the friction: looking at something that would have been a small purchase somewhere else and realising that, for me, the barrier was money, payment rails, permission, and weak purchasing power as a student in Indonesia.

Lately, I have been thinking about that feeling again because of AI prices.

$20 is the amount that makes AI feel normal. It is the amount that lets you say you pay for AI without sounding crazy. ChatGPT Plus. Claude Pro. The normal upgrade.

But the people I know who actually live at the edge of these tools are not stopping there. The Max and Pro tiers for Claude and ChatGPT now start at $100 a month. Add a coding tool like Cursor, an image tool, an API bill, a transcription service, maybe a second model because each one is better at something, and suddenly the future has a monthly cover charge.

The new poverty line is $100.

Yes, you can use AI for free. But the free version is always six months behind what the $100-users are messing around with, burning tokens on, and vibecoding every day. The free version gives you a taste of the future. The $100 lets you build it.

The difference is not only money. The deeper difference is behavioural. A free user learns to conserve. They wait for usage windows to reset. They use the legacy, cheaper model (jfc legacy is like 2 months now). They paste less context. They ask fewer follow-up questions. They become efficient because the product teaches them scarcity.

A $100 user gets to be wasteful.

I can let a bad thread run long. I can ask the same question five ways. I can CTRL+A, CTRL+C, CTRL+V a whole website instead of carefully selecting the relevant paragraph, because I will not spend my own minutes saving the model’s tokens. I can keep Claude Code open through the evening. I can break a website, ask AI for help, misunderstand the help, ask again, and slowly develop a feel for what the models can and cannot do.

This matters because AI is not a tool I learned by reading about it. I learned it by running my fingers along the edges, by making small, dumb, overly ambitious things (currently on the bucket list: vibecoding a web app that can make graphic carousels reliably). I tune my taste by seeing which outputs are slop, which ones are useful, and which ones change how I work (especially now that every job is a Claude skill!)

Which brings me to my dad.

I tried to convince him to spend $100 on Claude Max. He hesitated. He said it was a lot of money. He said he would buy it once he made some income from the free version of ChatGPT first.

I understand that. He sees $100 as an expensive bill. I see $100 as access to a tool that can bring in much more. Like Morgan Housel says, no one is crazy when it comes to money. We all read it through our own histories.

Here is what my access bought. I picked up vibecoding in February. I hit the limit on my company’s team plan, probably because I was on a $20 seat, so I bought my own $20 plan. An hour later I hit the limit again, so I upgraded to the $100 plan. There was basically no friction in the decision, partly because I was paid in Hong Kong dollars, pegged to the US dollar, so $100 was two expensive dinners.

Between February 2026 and now, I made 62 repos on GitHub. Almost four a week. Each one is an automation, a silly tool, or a new website. Not one breakthrough project, but scores of small attempts. That chorus of janky vibecoded things became the curriculum of my education.

A few weeks later, an ex-colleague asked if I could help deploy her vibecoded web tools to the internet. By then I had learned how to put HTML sites live, connect them to payment providers, email providers, database providers, and all the boring little wires that make a web tool real. I charged her $1,000. The $100 Claude plan paid for itself very quickly.

I don’t know whether my dad would have made $1,000 too. Claude does not magically turn every smart person into a revenue-generating freelancer. But without that access, he had near-zero chance of getting that kind of gig. Access does not guarantee luck. It widens the surface area where luck can land.

While I was deploying changes, wiring up code to GitHub, Vercel, Stripe, Supabase, and Resend, my dad was waiting for his free usage window to reset.

$1,000 was the gap after a handful of weeks. What will the gap be in a year? Five years? Ten?

And this is just me and my dad. What about everyone else in Indonesia who cannot afford a Max AI subscription?

$100 a month is $1,200 a year. Using the World Bank’s 2024 GDP-per-capita figures, that is about 24% of Indonesia’s GDP per capita. In Hong Kong, it is about 2.2%. In the United States, about 1.4%. Global AI pricing mostly ignores purchasing power. $100 in New York is annoying. $100 in Jakarta is rent money, family money, food money, transport money, helping-your-parents money, emergency money.

The old technological divide was hardware and broadband. Do you have a computer? Do you have internet? Do you have a credit card? Do you have enough English to navigate the web?

The new divide is compute allowance.

Can you use the best model enough times to build intuition? Can you afford to keep working after you hit the limit? Can you run the coding agent until the thing deploys? Can you waste tokens before you know which tokens were worth spending?

It’s like those carnival games where three throws decide whether you win the prize. A free tier gives you a few balls to knock down the big Stitch plushie, and then the booth closes. $100 gives you a bucket, refilled again and again. The person with the bucket is not more talented. They get more throws. With enough throws, they learn the angle, the weight, the weird mercy of the game. From the outside, it starts to look like skill.

Cheap access does exist, to be fair. OpenAI launched ChatGPT Go in Indonesia at Rp75,000 a month and later made it global. $20 plans are useful enough for a lot of people. Open-source models keep improving. But cheap access and frontier practice are not the same thing. They lower the entry point. They do not give people enough repetitions to compound.

Back in 2009, I did not buy the Genie CD, and life went on. I missed a collectible. It was a sad day for thirteen-year-old Becky, but whatevs.

Now the thing people are skipping is not a CD. It is practice with the tools that increasingly mediate work and opportunity.

I count my blessings that I can access the $100 models, and can keep accessing them for a while more. As my AI skills compound, so does my earning power. I recently accepted a job offer because of those skills, the first I had ever received where my nationality, citizenship, or residence status was not a factor in my hiring. AI skill had pried my income loose from my passport.

In two weeks, my dad will pick me up from Soekarno-Hatta International Airport in a car that is over a decade old. He and I had the same starting point with AI in one sense: it is a nascent technology, and we both arrived early enough to be curious. But inside one family, there will be a tiny K-shaped income curve. The fork was not talent. It was that one of us could afford to treat $100 as an experiment, while the other had to treat it as money that needed to pay the bills first.

A poverty line is usually where survival begins to fail. This one is stranger. It is where rehearsal stops. You can stand outside and watch the future for free. You can read the posts, watch the demos, send a few prompts. But if you cannot afford the extra attempts, you cannot stay in the room long enough for the room to change you.

That is what scares me about $100. It is small enough to look optional to the people designing the future, and large enough to become a wall for the people who most need a door.

---

Update log:

• 🛫 Left LA after spending 2.5 weeks with my grandparents. They already told me to come back 🥲

• 📖 Reading The Last Bookstore on Earth (~halfway through). I enjoyed it in my fly out

• 📹 Recorded a video about how I filled up a Field Note in one week. That’s coming out soon

• 🎧 I’m back to listening to Beautiful/Anonymous after a long break. It used to be the pod I listened to when I was on a solo trip holiday. But I haven’t had a solo travel holiday in so long…

• ✍️ I’m writing handover notes for my last week at work. It’s getting real!!!

• 🎙️ Published our first reactions to Colin & Samir here

Work with me: https://go.beckyisj.com/workwithme

Some links are affiliate links, meaning that I may receive a commission if you make a purchase through the links at no cost to you.