Pieter Levels: Why Indie Hacking Is Dead, AI Has No Moat, and the Platform Risk Nobody Talks About

Indie hacking is dead. At least, that’s what Pieter Levels once tweeted. Of course, it isn’t dead — but it is different now. And just how different, and why, is worth understanding.

Levels has been building startups in public for years: Nomad List, Remote OK, and more recently a pair of AI businesses that have generated serious revenue. He’s watched the indie hacking movement go from a subculture of misfits into the default way a generation approaches entrepreneurship. And he’s watched the ground shift underneath it.

Here’s what he sees.

‘Indie Hacking Is Dead’ — What He Actually Meant

The tweet was a long one. The first line was the provocative bit. Nobody read the second.

“What I meant was, indie hacking started around 2014 with Product Hunt, maybe 2016 properly. Before that, a lot of people would do the VC route. Then it became very popular and a lot of people are indie hacking.”

And now the big tech companies are watching. “Big tech companies follow me and they follow other indie makers to see what indie makers are doing. It’s finally on the radar of tech.”

He points to a tweet from someone in San Francisco living in a hostel for $600 a month, full of indie hackers trying to ship AI startups. “Why? Because it’s San Francisco, the place where you usually raise VC. So it means indie hacking has become a mainstream term.”

Which means more competition. More saturation. And you’re no longer competing only with other indie hackers — you’re competing with companies that raised $500 million.

“If remote work was called remote hacking, and then COVID happens and remote work becomes normal — it would also be dead, because it’s not hacking anymore. It’s just normal now. That’s exactly the difference.”

Indie hacking is just indie business. Indie entrepreneurship. It arrived.

The Recession Makes It a Necessity

There’s a silver lining. When money dries up, lean becomes mandatory.

“The economic recession is a very good point, because you want to do more lean. Indie hacking becomes a necessity because there is no money. It’s hard to get funding. Seed funding, Series A — money has dried up in many ways, and people are much more selective in what they fund.”

He also notes that a lot of VC-funded founder friends now say their next startup will preferably be indie. Burnout is real.

The AI Moat Problem

Levels recently said the F-word: no moat.

“Everybody talks about it, even VC-funded AI founders. We DM and everybody says the same: retention is separate, churn is very high, and defensibility is a problem. You make something and immediately you have a lot of clones because everybody’s working with the same stuff.”

Everyone uses GPT-4 for text, or open models, or Stable Diffusion for images. The technology isn’t proprietary. What’s proprietary is how you combine the models and build the product around them. But a smart competitor can figure that out in three months. A less smart one, six months to a year.

“It’s very difficult to make an AI startup now.”

It cuts both ways. For customers, it’s excellent — variety, competition, downward pressure on price. For founders, it’s brutal. “Pure economics: the profit margin goes to zero.”

And his margins aren’t high. GPU costs eat them alive.

The GPU Bottleneck

The plan was that GPU costs would fall. They didn’t, because Nvidia can’t produce enough chips. “People are fighting over GPUs. It’s insane.”

So you end up layered in dependencies: the chip, the model provider, the API. Every layer is somebody else’s business.

The Price Hike That Felt Like a Scam

He tells a story. He built Avatar AI — one of the first big AI avatar startups. It went viral. He made several hundred thousand dollars in a month or two.

The fine-tuning service he relied on cost $3 per training run. Then the provider saw his tweets about revenue and raised the price to $20. He was selling the output for around $25.

“For like a month, they got all the money. And I couldn’t switch — they said it was difficult. Maybe they were telling the truth, they had problems getting GPUs too. But they increased the price and I felt kind of scammed.”

He compares it to coworking spaces that raise rent tenfold once a tenant is successful, and to landlords generally.

He switched to Replicate, which he praises: helpful, no price increases, only decreases.

The lesson: “Think about alternatives from the start. If you have a service that you’re building on, abstract it enough so you can have another service to plug into your system.”

Ideally you run your own GPU server. But that means learning Python, and he’s honest about his limits. “I hired an AI developer for this. This year I see my limits. I’m not so good at this stuff.”

Hiring, and the Thing He Couldn’t Do Alone

He admits he was complacent. He thought he could do everything himself — front end, back end, everything — because Nomad List worked.

“This year I was like, man, there’s so many things I cannot do. Like marketing. I always rely on organic SEO.”

He points to Danny Postma (HeadshotPro) as someone who shook him up. Postma is an SEO machine — building businesses on top of keyword data, treating domains almost like real estate.

And then TikTok. “I hired a guy to help me because TikTok is insane. One influencer posted about Photo AI and MRR went from $12k to $40–50k. It stayed that way.”

Press, by contrast, does almost nothing. “You can put the logos on your site, but nobody links and clicks. It’s insane.”

The Tech Stack Meme

Levels builds in PHP and a bit of JavaScript. It’s become a meme.

“I exaggerated the meme for viral effect. People really think my code is bad. It’s not great, but it’s pretty good. It’s very clear, highly commented, multiple files, GitHub, a structural pattern I made myself, scheduled workers, SQLite.”

A PHP developer can get into it. He’s had developers come on for small things and find what they need.

But the meme has a point. “It’s not about PHP and jQuery. It’s about the fact that you have these developers who work in enterprise and agencies, and there’s this agency economy where a company that doesn’t know anything about tech comes to a web agency and wants a website. The agency needs to sell the best stuff, so they say we use the newest technology. Then developers need to update their skills. It’s a whole economy. Frameworks get funded, they have evangelists, they make developers promote the stuff.”

That makes new developers think they need all this technology. Often it makes things more complicated.

He invokes the Lindy effect: technology that has survived a long time tends to keep surviving. PHP is old and it just works. New technology breaks. “I go into an Airbnb and there’s a smart TV and it’s so difficult to watch TV now. Old TV just works.”

“If you’re an entrepreneur, it doesn’t matter what you use. There’s no need for this cultism. The problem is a lot of freelancers want to be entrepreneurs, and they bring the baggage of having to use this stack and over-engineer. The priority should be getting customers and getting people to pay money , because then you survive. You pay your rent.”

The Manual Concierge Approach

When Avatar AI launched, there was no code. Just an HTML page, examples of the avatars, input photos, a Typeform link, and a Stripe payment link.

“I would go to Stripe and check the email. On the Stripe checkout I had a link to a Typeform where it collected all the photos with file uploads. And then I would manually do the thing for 100 or 200 orders myself. Download the photos, go to the platform, upload, download the results. It was horrible manual work. I’d spend all night doing this. Then I started automating. After a week it was automatic.”

So: a landing page, a tweet, and a payment link. The code came after it worked.

He cites the XKCD cartoon about automation: the time to automate often exceeds the time the task itself takes. “So just do it yourself manually until it takes too much time.”

At $30 per photo, a night of manual work was $3,000. Not a bad night.

He also explains why he’s traumatized into this approach. “I spent like a year on a YouTube analytics startup in 2014. Nobody paid. The interface was amazing. But nobody wanted to be a customer. I’m so traumatized by this. I’m not going to build anything until there’s customers.”

And he has a tell for unvalidated products: too-beautiful landing pages. “Beautiful gradients, borders that move. If it’s too fancy, you spent too much time on design. I prefer a very ugly webpage in the beginning. Look at Google, look at Facebook — very ugly at the start. You need a very ugly, basic page to validate something first.”

What He’d Tell Someone Starting Today

“I think you should always do the opposite of what everybody else is doing. If indie hacking is mentioned, do something completely different. Go where nobody’s going. When I started, it was almost nobody — Patio11, Patrick McKenzie was bootstrapping. Everybody else was raising VC. Now it’s normal. So run away from the herd. Look for the part of the grass where nobody is.”

On social proof: “Maybe it’s not important. Maybe it’s important that you make something for a problem that’s out there.”

Where to find problems? Subreddits — every subreddit has a goal, and you can build around it. Craigslist — every category has become a startup. And now TikTok.

“Do go to TikTok and see what’s going on there. And document your journey — not on Twitter, but on TikTok. ‘Hey, I’m Pieter, I’m just starting out, I want to make some money and pay my rent.’ Every day make a video about what you’re doing and this new feature. That’s how I would approach it in 2023. I probably wouldn’t be on Twitter at all.”

The Generational Divide on AI

His take on the AI doom conversation is blunt. Gen Z doesn’t care — they just use it. He points to Jenni AI, an app students use to help write papers, reportedly at $150k MRR.

“Nobody says, is this good or bad? That question comes from millennials. We’re millennials. Journalists are millennials. Maybe it’s our culture to challenge everything, which is kind of good, and to make a problem out of everything.”

There’s a real foundation to the concern, he admits. “I understand completely that if you cannot code and you see this AI stuff happening, and your income is not increasing because incomes are stagnant, and the government is not providing basic income — you feel bad. You’re going to complain.”

His answer is policy, not panic. “I think the government should provide basic income to most people. I don’t know if there will be a lot of new jobs created — I don’t really believe in that. But the goal for everybody in humanity should be to live a good life. That should be protected, so that technology doesn’t destroy people’s income.”

Nomad List, Nine Years Later

Against the AI churn sits Nomad List — nine years old, averaging $40–60k a month, low costs, stable.

“I’d never expect to stay so long. I had a YouTube channel before — electronic music mixes — that went up really fast, $8k a month, then went down very fast. So I thought every business is hype. I was traumatized. But it keeps going.”

He’s currently building a 3D globe for it — lines tracing his travel from Thailand to Qatar to Holland to Brazil. “It’s my baby. You can work on this until you’re 80, because there’s always a different way to figure out the best place to live.”

The contrast with AI is stark. “This AI stuff is so stressful because you always need to stay ahead of competitors. Last year every week there was some new breakthrough and you need to implement it very fast. It fucks with your sleep.”

Would he sell? “I think I’ll get it to a certain level, maybe $100k MRR for Photo AI and Interior AI, and then — I got them valued recently, and the multiples for AI startups are very good. Normally 2–3x, but for AI it can be 5, 6, even 8 because it’s hype. The problem is profit — multiples are based on profits, so you need to cut costs rapidly.”

Twitter After the Change

The algorithm changed. “Before, everything would get views. Now if something goes viral in the beginning, it gets pumped maximum — 1,000 retweets. But on the other side, many tweets get two likes. I have 300,000 followers. It’s bizarre.”

He thinks the platform now measures watch time in the first seconds and only amplifies tweets that perform. “When a tweet doesn’t come from an extreme place, they just don’t show it. So it becomes more extreme. Tweets have to go really far to get retweets, or they get nothing.”

Product updates, which used to do well, now barely register. “The long-term vibe of Twitter changed a little bit.”

But he keeps tweeting. “If nobody cares anymore, I’ll keep tweeting, because I was tweeting and nobody cared ten years ago.”

On Being Yourself

“Don’t be trend-boy. Don’t do ‘the five AI tools you need in 2024, number three will shock you.’ It’s bullshit.”

What works instead: long posts, which get view time, and video. “Now you need YouTube ideas on Twitter. The intro, the MrBeast style, thumbnails. It’s different. But that’s the same sentiment as indie hacking dead, nomadism dead, Twitter dead. It’s all just different.”

And on the trolls: “I think it’s really funny. But you need to see it as a game. It’s impossible to see it as a normal conversation. If you have 300,000 people on the other side, it’s just shouting. It’s like a lynch mob. That changes you, and you have to watch out that it doesn’t change your personality in real life.”

He’s honest about changing his mind. “I have strong opinions, weakly held. I change my opinions all the time. I like to say something and then hear what people reply. I learn from that.”

“The reason a lot of people follow me is because they know I’m honest and I’m not perfect. I write whatever I think. I don’t want to become fake.”

And the cost of that: “People are getting really angry and they hate you for your opinions. I saw four people in three days saying I need a break from Twitter. It’s unfortunate.”

His final advice isn’t complicated. Don’t be angry about the changes. Embrace them and reinvent yourself for the new cycle. Look for the grass where nobody else is standing. Build the ugly page first. Talk to the people who actually have the problem.

And keep tweeting even when nobody’s watching, because that’s what you were doing anyway.