Interview with the founder of OpenClaw: AI is a lever, not a substitute; 80% of apps will be replaced.

Author: Baoyu

This is another 40-minute interview with Peter Steinberger, author of ClawdBot/OpenClaw, hosted by Peter Yang.

Peter is the founder of PSPDFKit and has nearly 20 years of experience developing for iOS. After receiving a €100 million strategic investment from Insight Partners in 2021, he chose to "retire." Now, his creation, Clawdbot (now renamed OpenClaw), has become incredibly popular. Clawbot is an AI assistant that can chat with you via WhatsApp, Telegram, and iMessage, and it's connected to various applications on your computer.

Peter described Clawbot as follows:

"It's like a friend living in your computer, a little weird, but frighteningly smart."

In this interview, he shared many interesting perspectives: why complex agent orchestration systems are "slop generators," why "making AI run 24 hours a day" is a vanity metric, and why programming languages ​​are no longer important.

One-hour prototype, 300,000 lines of code

Peter Yang asked him what Clawbot was and why its logo was a lobster.

Peter Steinberger didn't directly answer the lobster question, but instead told a story. After "retiring," he devoted himself entirely to vibe coding—the kind of work where an AI agent writes code for you. The problem is, the agent might run for half an hour, or it might stop and ask you a question after just two minutes. You go out for a meal and come back to find it's completely stuck, which is very annoying.

He wanted something that would allow him to check his computer's status anytime on his phone. But he didn't pursue it because he felt it was too obvious; big companies would definitely do it.

"When no one had done it by November of last year, I thought, forget it, I'll do it myself."

The initial version was extremely simple: connect WhatsApp to Claude Code. Send a message, and it uses AI to send back the result. It was set up in just one hour.

Then it "came back to life." Now Clawbot has about 300,000 lines of code and supports almost all major messaging platforms.

"I think this is the direction of the future. Everyone will have a super powerful AI that will accompany you throughout your life."

He said, "Once you give AI access to your computer, it can basically do anything you can."

That morning in Morocco

Peter Yang said that now you don't need to sit in front of the computer and stare at it anymore; you can just give it instructions.

Peter Steinberger nodded, but he wanted to talk about something else.

Once, while celebrating a friend's birthday in Morocco, he discovered he had been using Clawbot all along. Asking for directions and finding restaurant recommendations were minor tasks. What truly surprised him was that morning: someone tweeted that one of his open-source libraries had a bug.

"I just took a picture of the tweet and sent it to WhatsApp."

The AI ​​understood the tweet's content and recognized it as a bug report. It checked out the corresponding Git repository, fixed the issue, committed the code, and then replied to the person on Twitter saying it was fixed.

"I thought to myself, 'This is possible?'"

And there was another even more absurd instance. He was walking down the street, too lazy to type, so he sent a voice message. The problem was, he hadn't provided any support for writing voice messages for Clawbot.

"When I saw it display 'typing,' I thought, 'This is it.' But it replied normally."

He later asked the AI ​​how it did it. The AI ​​said: "I received a file but it didn't have an extension, so I looked at the file header and found it was in Ogg Opus format. You have ffmpeg on your computer, so I used it to convert it to WAV. Then I looked for whisper.cpp, but you didn't have it installed. However, I found your OpenAI API key, so I used curl to send the audio over and transcribed it."

After listening, Peter Yang said: These things are really ingenious, although they are a bit scary.

"It's so much better than the web version of ChatGPT; it's like ChatGPT has been unleashed. Many people don't realize that tools like Claude Code aren't just great for programming; they can solve any problem."

Command Line Tool (CLI) Legion

Peter Yang asked him how those automated tools were built, whether he wrote them himself or had AI write them.

Peter Steinberger laughed.

For the past few months, he's been expanding his "CLI army." What are intelligent agents best at? Calling command-line tools, because that's all there is in the training data.

He built a CLI to access all Google services, including the Places API. He also created a tool specifically for searching for emojis and GIFs so that the AI ​​could send memes when replying to messages. He even made a tool to visualize sound, hoping to let the AI ​​"experience" music.

"I also hacked into the API of a local food delivery platform, and now AI can tell me how long it will take for my food to arrive. I also reverse-engineered the Eight Sleep API so I can control the temperature of my bed."

[Note: Eight Sleep is a smart mattress that can adjust the surface temperature; its API is not officially available.]

Peter Yang followed up: Did you have AI build all of this for you?

"The most interesting thing is that I previously worked at PSPDFKit for 20 years developing for the Apple ecosystem, and I was very proficient in Swift and Objective-C. But after coming back, I decided to switch tracks because I was fed up with Apple interfering in everything, and the audience for Mac apps was too narrow."

The problem is, switching from a technology stack you're proficient in to another is a painful process. You understand all the concepts, but not the syntax. What is a prop? How do you split an array? You have to look up every little question, and you'll feel like an idiot.

"Then came AI, and all of that disappeared. Your systems thinking, architectural skills, taste, and judgment of dependencies—these are what truly have value, and now they can be easily transferred to any field."

He paused for a moment:

Suddenly I felt like I could build anything. Language didn't matter anymore; what mattered was my engineering mindset.

Control the real world

Peter Steinberger began demonstrating his setup. The list of permissions he granted to the AI ​​was staggering:

Emails, calendar, all documents, Philips Hue lights, Sonos speakers. He can have the AI ​​wake him up in the morning and gradually increase the volume. The AI ​​can also access his security cameras.

“Once I had it keep an eye out for strangers. The next morning it told me, ‘Peter, someone’s here.’ When I checked the video, it had been taking screenshots of my sofa all night. Because the camera’s image quality was poor, the sofa looked like someone was sitting on it.”

In the Vienna apartment, AI can also control the KNX smart home system.

"It really can lock me out."

Peter Yang asked: How were these connected?

"Just tell it directly. These things are very resourceful; it will find APIs, it will Google, and it will look for keys in your system."

Users are getting even crazier:

• Someone told it to buy from Tesco's online store.
• Someone ordered it from Amazon.
• Someone set it to automatically reply to all messages.
• Someone added it to the family group chat as a "family member".

"I had it check me in on the British Airways website. It was practically a Turing test; you know how counterintuitive the interface is when you're trying to navigate a browser on an airline's website."

The first time it took almost 20 minutes because the whole system was still quite rudimentary. The AI ​​needed to find its passport in its Dropbox, extract the information, fill out a form, and pass the human verification.

"Now it only takes a few minutes. It can click the 'I am human' verification button because it is controlling a real browser, and its behavior is no different from that of a human."

80% of apps will disappear.

Peter Yang asked: What are some safe and easy-to-use tips for regular users who have just downloaded the app?

Peter Steinberger said everyone's path is different. Some people install it and immediately start writing iOS apps, while others immediately start managing Cloudflare. One user installed it for himself in the first week, for his family in the second week, and started building an enterprise version for his company in the third week.

"After I installed it for a non-technical friend, he started sending me pull requests. He had never sent a pull request in his life."

But what he really wanted to convey was a bigger picture:

"If you think about it, this thing could replace 80% of the apps on your phone."

Why would you still use MyFitnessPal to track your diet?

“I have an infinitely resourceful assistant that already knows I made the wrong decision at KFC. I post a photo, and it saves it to its database, calculates the calories, and reminds me to go to the gym.”

Why bother setting the temperature for Eight Sleep using an app? AI has API permissions and can adjust it for you directly. Why bother using a to-do list app? AI will keep track for you. Why bother using an app to check in flights? AI will do it for you. Why bother using a shopping app? AI can recommend products, place orders, and track transactions.

“An entire layer of apps will slowly disappear because if they have APIs, they are just services that your AI will call.”

He predicts that 2026 will be the year many people begin to explore personal AI assistants, and large companies will also enter the market.

"Clawbot may not be the ultimate winner, but it's on the right track."

Just Talk to It

The topic then shifted to AI programming methodologies. Peter Yang mentioned writing a very popular article called "Just Talk to It," and asked if he could elaborate on it.

Peter Steinberger's core argument is: Don't fall into the "agentic trap".

“I see too many people on Twitter discovering that intelligent agents are powerful, then trying to make them even more powerful, and then falling down the rabbit hole. They build all sorts of complex tools to speed up workflows, but they're just building tools, not anything of real value.”

He himself had fallen into this trap. Early on, he spent two months building a VPN tunnel just to access the terminal on his phone. He did it so well that once, while having dinner with friends, he spent the entire time coding on his phone using Vibe instead of participating in the conversation.

"I had to stop, mainly for mental health reasons."

Slop Town

What's been driving him crazy lately is an orchestration system called Gastown.

"An extremely complex orchestrator running a dozen or twenty agents simultaneously, communicating and dividing tasks among themselves. There are watchers, overseers, mayors, pcats (which may refer to 'civilians' or 'pet cats' as filler roles), and I don't even know what else there are."

Peter Yang: Wait, there's also the mayor?

"Yes, there is a mayor in the Gastown project. I call this project 'Slop Town'."

There's also the RALPH mode (a "use-and-play" single-task loop mode where the AI ​​is given a small task, completed, and then all contextual memories are discarded, everything is reset and started over, resulting in an infinite loop)...

"This is the ultimate token burner. You let it run all night, and the next morning you get the ultimate slop."

The crux of the problem is: these intelligent agents lack taste. They are frighteningly intelligent in some ways, but if you don't guide them, don't tell them what you want, all that comes out is garbage.

"I don't know how others work, but when I start a project, I only have a vague idea. As I build, play with, and experience, my vision gradually becomes clearer. I try some things, some don't work, and then my idea evolves into its final form. My next prompt depends on the current state of being that I see, feel, and think about."

If you try to write everything into the early specifications, you miss out on this human-machine loop.

"I don't know how you can make something good without feeling and appreciating the process."

Someone on Twitter was showing off a note-taking app that was "fully RALPH generated." Peter replied: Yes, it looks like it was generated by RALPH; no normal person would design it like that.

Peter Yang concludes: Many people run AI for 24 hours not to develop an app, but to prove that they can make AI run for 24 hours.

"It's like a size contest with no reference point. I've run loops for 26 hours, and I was quite proud of it at the time. But that's a vanity metric, meaningless. Being able to build everything doesn't mean you should build everything, nor does it mean it will be good."

Plan Mode is a patchwork (hack).

Peter Yang asked him how to manage the context. If the conversation gets too long, the AI ​​might get confused; should it be manually condensed or summarized?

Peter Steinberger said this was "a problem with the old model."

"Claude Code still has this problem, but Codex is much better. On paper, it may only have 30% more context, but it feels like 2-3 times more. I think it's related to the internal thinking mechanism. Now most of my feature development can be completed within a single context window, with discussion and building happening simultaneously."

He doesn't use Worktrees because he sees it as "unnecessary complexity." He simply checks out several repositories: clawbot-1, clawbot-2, clawbot-3, clawbot-4, and clawbot-5. He uses whichever is available, completes the tests, pushes it to the main branch, and then synchronizes.

"It's a bit like a factory, if they're all busy. But if you only have one open, the waiting time is too long to get into a flow state."

Peter Yang said it's like a real-time strategy game where you have a team attacking and you have to manage and monitor them.

Regarding the plan mode, Peter Steinberger has a controversial viewpoint:

"Plan mode was a patchwork solution that Anthropic had to add because the model was too impulsive, jumping straight into writing code. If you're using the latest model, like GPT 5.2, you're essentially talking to it. 'I want to build this feature, it should be like this or that, I like this design style, give me a few suggestions, let's discuss them first.' Then it will make suggestions, you discuss them, reach a consensus, and then take action."

He doesn't type, he speaks.

I talk to it most of the time.

Discord driver development

Peter Yang asked him what his process was for developing new features. Was it to first explore the problem? Or to first make a plan?

Peter Steinberger said he did something that was "probably the craziest thing I've ever done": he connected his Clawbot to a public Discord server, allowing everyone to talk to his private AI, carrying his private memories, in public.

"This project is hard to describe in words. It's like a hybrid of Jarvis (the AI ​​assistant in Iron Man) and the movie Her. Everyone I demonstrated it to in person was super excited, but it just didn't go viral when I posted pictures and captions on Twitter. So I thought, why not let people experience it for themselves?"

Users ask questions, report bugs, and make requests on Discord. His current development process is: take a screenshot of a Discord conversation, drag it into the terminal, and say to the AI, "Let's talk about this."

"I'm too lazy to type. When someone asks, 'Do you support this or that?' I just have AI read the code and write an FAQ."

He also wrote a web crawler that scans Discord's help channel at least once a day, allowing AI to summarize the biggest pain points, which they then fix.

No MCP, no complex orchestration

Peter Yang asked: Do you use those fancy things? Multi-agent systems, complex skills, MCP (Model Context Protocol) and the like?

"My skills are mostly life skills: recording meals, grocery shopping, and stuff like that. I have very little programming knowledge because I don't need it. I don't use MCP or anything like that."

He doesn't believe in complex orchestration systems.

"I'm in a loop, and I can make products that feel better. There may be faster ways, but I'm almost at the point where the bottleneck isn't AI anymore. I'm mainly limited by my own thinking speed, and occasionally by the time I have to wait for Codex."

His former PSPDFKit co-founder, a former lawyer, is now also sending him PRs (pull requests).

"AI allows people without a technical background to build things, which is amazing. I know some people will object and say that this code is not perfect. But I think of pull requests as prompt requests; they convey intent. Most people don't have the same understanding of the system and can't guide the model to the optimal result. So I prefer to get the intent and do it myself, or rewrite it based on their PR."

He would mark them as co-authors, but rarely directly merge other people's code.

Find your own path

Peter Yang concludes: So the key point is, don't use a slop generator, keep people in the loop, because the human brain and taste are irreplaceable.

Peter Steinberger added:

"Or rather, find your own path. Many people ask me, 'How did you do it?' The answer is: you have to explore it yourself. Learning these things takes time and requires making your own mistakes. It's the same as learning anything, except that this field changes particularly rapidly."

Clawdbot can be found on clawd.bot and GitHub. Clad with a W, CLAWDBOT, looks like a lobster claw.

(Note: ClawdBot has been renamed OpenClaw)

Peter Yang said he had to give it a try. He didn't want to sit in front of the computer chatting with AI; he wanted to be able to give it commands anytime while he was out with his child.

"I think you'll like it," Peter Steinberger said.

Peter Steinberger's core argument can be summarized in two sentences:

1. AI has become so powerful that it can replace 80% of the apps on your phone.
2. But without human taste and judgment in the cycle, the output is just garbage.

These two statements, seemingly contradictory, actually point to the same conclusion: AI is a lever, not a substitute. It amplifies what you already possess: systems thinking, architectural capabilities, and an intuition for good products. Without these, running multiple intelligent agents in parallel for 24 hours a day is merely mass-producing SLOPS (Scaling Loops).

His practice itself is the best proof: a 20-year veteran iOS programmer built a 300,000-line code project in TypeScript in a few months, not by learning the syntax of a new language, but by those language-independent things.

"Programming languages ​​are no longer important; what matters is my engineering mindset."