Clawdbot Founder Interview: AI is Leverage, Not Replacement

Reposted from Baoyu (@dotey)’s Twitter, summarizing a 40-minute interview with Clawdbot/OpenClaw creator Peter Steinberger by Peter Yang.

Peter Steinberger is the founder of PSPDFKit with nearly 20 years of iOS development experience. After his company received a €100 million strategic investment from Insight Partners in 2021, he chose to “retire.” Now, his creation Clawdbot (renamed OpenClaw) has exploded in popularity—an AI assistant that chats with you via WhatsApp, Telegram, or iMessage, connected to all the applications on your computer.

Peter describes Clawdbot like this:

“It’s like a friend living in your computer. A bit weird, but scary smart.”

One-Hour Prototype, 300,000 Lines of Code

After coming back from “retirement,” Peter went all-in on vibe coding—having AI agents write code for you. The problem is, agents might run for 30 minutes or stop after 2 minutes to ask a question. You go eat lunch and come back to find it’s been stuck for hours.

He wanted something to check his computer’s status from his phone anytime. But he didn’t build it, thinking big companies would surely do it.

“By last November when nobody had, I thought, fine, I’ll do it myself.”

The initial version was dead simple: connect WhatsApp to Claude Code. Send a message, it calls the AI, sends back the result. Built in one hour.

Then it “came alive.” Now Clawdbot has about 300,000 lines of code and supports almost every major messaging platform.

“I think this is the direction of the future. Everyone will have a super powerful AI that follows them through life.”

That Morning in Morocco

Once Peter was in Morocco for a friend’s birthday. Someone tweeted that one of his open-source libraries had a bug.

“I just took a photo of the tweet and sent it to WhatsApp.”

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

“I thought, wait, that works?”

Another time was even crazier. He was walking down the street, too lazy to type, so he sent a voice message. The thing is, he had never built voice message support into Clawdbot.

“I saw ’typing…’ and thought, well, this is going to break. Then it replied normally.”

He later asked the AI how it did it. The AI said: I received a file with no extension, so I looked at the file header and found it was Ogg Opus format. You have ffmpeg on your computer, so I converted it to WAV. Then I looked for whisper.cpp but you don’t have it, but I found your OpenAI API key, so I used curl to send the audio for transcription.

“Way better than web ChatGPT. This is like ChatGPT unchained.”

The CLI Army

Over the past few months, Peter has been building his “CLI army.” What are agents best at? Calling command-line tools, because that’s what the training data is full of.

He built:

• A CLI for all Google services including Places API
• A meme and GIF finder so AI can send memes when replying
• A sound visualization tool so AI can “experience” music
• Hacked into local food delivery platform’s API so AI can tell him when food will arrive
• Reverse-engineered Eight Sleep’s API to control his bed temperature

“I spent 20 years doing Apple ecosystem development at PSPDFKit—Swift, Objective-C, very specialized. But when I came back, I decided to switch tracks.”

The problem is, switching from one mastered tech stack to another is painful. You know all the concepts but not the syntax. Every little thing requires googling. You feel like an idiot.

“Then AI came along and all that disappeared. Your systems-level thinking, architecture skills, taste, judgment about dependencies—those are what’s truly valuable, and now they transfer easily to any domain.”

“Suddenly I felt like I could build anything. The language doesn’t matter anymore. What matters is my engineering mindset.”

Controlling the Real World

Peter’s permission list for his AI is jaw-dropping:

Email, calendar, all files, Philips Hue lights, Sonos speakers, security cameras, KNX smart home system.

“It can literally lock me out of my house.”

Users’ use cases are even crazier:

• Someone has it do Tesco grocery shopping
• Someone has it place Amazon orders
• Someone has it auto-reply to all messages
• Someone added it to a family group chat as a “family member”

“I had it check me in on British Airways website. It can click the ‘I am human’ verification button because it’s controlling a real browser with behavior patterns identical to a human.”

80% of Apps Will Disappear

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

Why use MyFitnessPal to log meals? Send a photo and AI will store it in a database, calculate calories, remind you to hit the gym.

Why use an app to set the temperature? AI has API access, just adjusts it directly.

Why use a to-do app? AI remembers for you.

Why use an app to check in for flights? AI does it for you.

“There’s a whole layer of apps that will slowly disappear, because if they have APIs, they’re just services your AI will call.”

Don’t Fall into the Agentic Trap

Peter’s core message: Don’t fall into the “agentic trap.”

“I see too many people on Twitter discover agents are powerful, then want to make them more powerful, then fall down the rabbit hole. They build complex tools to speed up workflows, but end up just building tools, not building anything truly valuable.”

What drives him crazy lately is an orchestration system called Gastown:

“A super complex orchestrator running a dozen or twenty agents simultaneously, communicating with each other, dividing work. Watchers, overseers, a mayor… I call this project ‘Slop Town.’”

And RALPH mode (fire-and-forget single-task loops):

“This is the ultimate token burner. You let it run overnight, and next morning you get ultimate slop.”

The core problem: These agents don’t have taste yet.

“I don’t know how you can make something good without feeling, without taste being involved.”

Peter Yang summarized: A lot of people run AI for 24 hours not to build apps, but to prove they can run AI for 24 hours.

“It’s like a comparison contest with no reference point. Being able to build everything doesn’t mean you should, and doesn’t mean it will be good.”

Plan Mode is a Hack

On plan mode, Peter has a controversial take:

“Plan mode is a hack Anthropic had to add because the model is too impulsive, it just jumps straight into coding. If you use the latest models like GPT 5.2, you just talk to it. Discuss requirements, debate approaches, reach consensus, then start building.”

He doesn’t type, he talks:

“I speak to it most of the time.”

No MCP, No Complex Orchestration

“My skills are mostly life skills: logging meals, grocery shopping, that kind of thing. Very few for programming, because I don’t need them. I don’t use MCP, I don’t use any of that stuff.”

He doesn’t believe in complex orchestration systems:

“I’m in the loop, I can make products that feel better. Maybe there’s a faster way, but I’m already so fast that the bottleneck isn’t AI anymore—I’m mainly limited by my own thinking speed.”

Core Takeaway

Peter Steinberger’s core message can be summarized in two sentences:

1. AI is now powerful enough to replace 80% of the apps on your phone
2. But without human taste and judgment in the loop, the output is slop

These two statements seem contradictory but point to the same conclusion: AI is leverage, not replacement. It amplifies what you already have: systems thinking, architecture skills, intuition for good products. If you don’t have these, running agents in parallel for 24 hours just mass-produces slop.

His own practice is the best proof: a 20-year iOS veteran built a 300,000-line TypeScript project in a few months, not by learning new language syntax, but by leveraging language-agnostic skills.

“Programming languages don’t matter anymore. What matters is my engineering mindset.”

Original summary by: Baoyu (@dotey)