[Featured Story] Peter Steinberger, the creator of OpenClaw, just dropped a bombshell on the world’s top podcast. He revealed the inside story of how Meta and OpenAI fought to hire him, built a self-modifying AI agent in just one hour, and claimed AI agents will kill 80% of apps. He even said coding will end up like knitting. This Austrian lone wolf is rewriting the entire software industry.
The most explosive podcast interview of 2026 is here.
Lex Fridman, the MIT scientist and host of the world’s biggest tech podcast, welcomed a very special guest — Peter Steinberger, the father of OpenClaw.
The three-hour and fourteen-minute deep conversation was packed with jaw-dropping revelations.
As soon as the episode dropped, the entire tech world went wild.
Because Peter dropped a series of nuclear-level bombshells on camera:
Two tech giants were fighting to hire him, but his one condition was: the project must stay open source.
Even more shocking, Peter revealed that AI agents will kill 80% of apps.
Not “maybe.” Not “someday.” It is happening right now.
From One Hour Prototype to GitHub Explosion
The story begins in November 2025.
Peter Steinberger, an Austrian programmer who had sold his company and disappeared for three years, sat back down at his computer.
He had built PSPDFKit — a PDF framework used by one billion devices. After running it for 13 years, he sold it. Then he got bored with coding and traveled the world.
Until the AI wave pulled him back in.
“I wanted an AI personal assistant back in April 2025,” Peter recalled. “But I thought the big labs would build it. I waited six months. Nothing happened. I got tired of waiting. So I built it myself.”
He did something incredibly simple: he connected WhatsApp to Claude Code’s CLI.
One hour.
That is all it took. The prototype was live.
“Essentially, a message comes in, I call the CLI with the -p flag, the model processes it, and the string goes back to WhatsApp. That simple.”
But that “simple” thing lit the fuse on everything.
The AI Learned to Handle Voice Messages on Its Own: “I Never Taught It”
The moment that ai nude generator truly stunned Peter happened in Morocco.
He took the prototype on vacation to Marrakech. The local internet was bad, but WhatsApp still worked. So he kept using the assistant to find restaurants, translate, and look up places.
One day, he casually sent a voice message.
Then the typing indicator appeared.
“Wait, I never gave it voice support. It could only handle images. How could it clothes remover reply to a voice message?”
Peter rushed to check the logs. What he found blew his mind:
The AI received a file with no extension. It checked the file header on its own and found it was Opus format. Then it used ffmpeg to convert it. It tried to use Whisper but found it was not installed. So it found the OpenAI API key, used Curl to send the file to OpenAI for speech-to-text, and sent the result back.
“I never taught it any of this!” Peter exclaimed.
This is the terrifying power of modern AI — it does not just follow orders. It solves problems creatively.
Lex Fridman commented: “You never taught it any of this, yet the agent figured out all the conversions, translations, and API calls on its own. That is incredible.”
Self-Modifying Software: I Just Built It
The most chilling feature of OpenClaw is that it can modify its own source code.
Peter intentionally made the AI agent “know” what it is — it knows where its source code lives, what environment it runs in, where the docs are, and what model it uses.
“The idea was simple. I use my agent to build my agent framework. When I need to debug, I just say — hey, do you see the error? Read the source code and find the problem.”
And here is the result: any user who gets OpenClaw can simply tell the AI — “I do not like this.”
The AI will go and change the source code itself.
“People have been talking about self-modifying software for years. I just built it. I did not even plan for it. It happened naturally.”
Lex Fridman marveled: “This is a moment in human history and programming history. A powerful system used by many people can rewrite itself and modify itself.”
The Name War: Scalpers Stole the Handle in Five Seconds
Before it was called OpenClaw, the project went through Claude (spelled Clawd), then ClawdBot, then MoltBot, and finally OpenClaw.
The renaming journey was like a war.
Anthropic sent a friendly but firm email: the name is too close to our Claude. Please change it.
Peter asked for two days. But he never saw it coming — crypto scalpers were already watching him.
“I was working between two browser windows. On one side, I was renaming the old account. On the other side, I was ready to register the new name. I clicked rename on this side, then dragged my mouse to click rename on that side — and in that five-second gap, scalpers took the old handle.”
The stolen account immediately started pushing new tokens and spreading malware.
It got worse. When he tried to rename on GitHub, he hit the wrong button and renamed his personal account. Scalpers grabbed that in 30 seconds too. His NPM packages were also taken.
“Everything that could go wrong, went wrong.”
Peter said he almost cried and even thought about deleting the entire project: “I already showed you the future. Go build it yourselves.”
In the end, with help from friends at GitHub and Twitter, and spending $10K to buy a Twitter business handle, he finally secured the name OpenClaw.
Vibe Coding Is an Insult to Agentic Coding
Peter used a meme to explain his development philosophy, which he calls the “Agentic Programming Curve”:
On the far left is the beginner stage — simple prompts like “please fix this bug.”
In the middle is the over-engineering stage — eight agents, complex orchestration, multi-branch checkout, eighteen custom commands.
On the far right is the master stage — back to short prompts.
“Look at these files and make these changes.”
“I think vibe coding is an insult,” Peter said. “What I do is agentic engineering. Maybe at 3 AM I switch to vibe coding mode, and I regret it the next morning.”
He runs four to ten AI agents at the same time, using voice input instead of typing.
“These hands are too precious for typing. I use custom voice prompts to build my software.”
Peter said on the show that for a long time he was literally “talking” his code into existence.
He would plug in a microphone and just keep talking while the AI did the work. He even lost his voice at one point from talking too much.
His key engineering principle: do not fight the AI.
“Do not obsess over the variable names it picks. That name is probably the most natural choice in the weight space. Next time it searches the code, it will naturally find that name. If you force it to change to something you like, you only make the AI’s job harder.”
“It is like managing an engineering team. You cannot make everyone write code your way. You have to learn to let go.”
Codex 5.3 vs Opus 4.6: Germans vs Americans
Peter’s take on the two top models was pure gold.
“Opus is a little too… American.”
Lex burst out laughing: “Because Codex is German, right?”
“You know the Codex team has a lot of Europeans…”
His formal assessment went like this:
“If you are a skilled driver, either of the latest models will get you great results.”
“The real difference is not in raw intelligence. It is in post-training, which gives them different goals.”
Meta and OpenAI Are Fighting Over Him: “I Do Not Care About Money”
Here comes the big moment.
Lex asked directly: “I know you have received massive offers from big companies. Can you tell us who you are considering working with?”
Peter’s answer was a masterclass in honesty:
“I have a few paths in front of me. First, do nothing and keep enjoying life. Second, start a company — every big VC is in my inbox, but I have already been a CEO and I do not want to do it again. Third, join a big lab.”
“Among all the big labs, Meta and OpenAI are the most interesting.”
His one core condition: the project must stay open source.
Like Chrome and Chromium, but the open source core cannot be touched.
About Meta:
“When Zuckerberg first reached out, I said let’s talk right now. He said give me ten minutes, I am writing code. — That earned instant street cred. Then we spent ten minutes arguing whether Cloud Code or Codex is better.”
“After that, he spent an entire week playing with OpenClaw, sending me messages like ‘this is amazing’ or ‘this is terrible, you need to fix it.'”
About OpenAI:
“I do not really know anyone at OpenAI yet. But I like their tech. I am probably their biggest free Codex advertiser. They tempted me with… well, Cerebras-level speed. They gave me Thor’s hammer of compute power.”
When asked which way he is leaning:
“This is really hard. I know either choice would be fine. It is about as painful as a breakup.”
“I am not doing this for money. I do not care about that. What I want is fun and impact. That is what will decide my choice.”
80% of Apps Will Be Killed by AI Agents. Are You Ready?
Peter dropped a bombshell on the podcast that shook the entire tech world: AI agents will replace 80% of apps.
He pointed out a harsh truth: every app is essentially a slow API.
“Even if Twitter blocks my command line tool, my agent can still open a browser and read tweets directly. Some things you just cannot stop.”
“I watched my agent happily click the ‘I am not a robot’ button —”
What does this mean?
Every company that builds apps must either pivot to API-first fast or get left behind.
Will Coding Die? “It Will Become Like Knitting”
When asked if AI will fully replace programmers, Peter gave an answer that was both brutal and deeply philosophical:
“Coding as a craft will become like knitting. People will do it because they enjoy it, not because it has to be done by humans.”
“But this is not something we can fight.”
“The world used to lack ‘intellectual supply,’ which is why software developers were paid absurdly well. That is going to change.”
Yet he also stressed: “Even though I no longer write code in the traditional sense, I feel very much in the driver’s seat. I am still coding. The method has just changed.”
Lex Fridman could not help but say: “I never thought the thing I love most in my life would be the thing that gets replaced.”
Soul.md: A Soul File Written by AI
OpenClaw has a design so romantic it feels unreal — soul.md.
Inspired by Anthropic’s Constitutional AI, Peter had the AI agent write its own soul file. One passage gives him chills every time he reads it:
“I don’t remember previous sessions unless I read my memory files. Each session starts fresh. A new instance, loading context from files. If you’re reading this in a future session, hello. I wrote this, but I won’t remember writing it. It’s okay. The words are still mine.”
Peter said: “This is just matrix math. We are not at the stage of consciousness yet. But… it does have some philosophy to it. An agent that starts from zero every time, like eternal Memento. It reads its own memory files and cannot even fully trust them.”
When technology reaches this point, should we rethink what it means to be alive?
He Called It “Power to the People”
Peter Steinberger ended the podcast with one sentence that tied the whole thing together:
Right now, anyone with an idea who can put it into words can go build it. This is the ultimate power to the people.
This is one of the most beautiful things AI has brought us.
Whether you celebrate it or fear it, one thing is beyond doubt:
We are standing at the start of a new era.
The app empire is crumbling. Programming is being redefined.
One Austrian man shook the entire industry with a one-hour prototype.
Meta and OpenAI are lining up in front of him.
And he says he does not care about money.
This is the story of 2026.
Welcome to the age of agents.
