How I Created OpenClaw, the Breakthrough AI Agent | Peter Steinberger | TED

0:04 What's OpenClaw?

0:06 I've been programming since I was 14.

0:11 Building software felt like playing a video game.

0:16 I couldn't stop.

0:19 Until I did.

0:23 I created a company, I poured a decade of my life into it.

0:29 No venture capital, every weekend.

0:34 And then I sold the dream.

0:38 And I felt absolutely nothing.

0:43 For three years, I was wondering,

0:48 did therapy, traveled, I changed countries twice.

0:55 Nothing, nothing clicked.

0:57 Now I wake up every morning with everything

1:01 I was supposed to want and no reason.

1:04 No reason to get out of bed.

1:07 And then in early 2025, I tried an experiment.

1:14 I wanted to see what these new AI coding agents are all about.

1:21 And I had what I can only describe as a "holy shit" moment.

1:28 The boilerplate, the plumbing,

1:31 all the boring parts that was software development, AI could do all of it.

1:39 The bottleneck is no longer typing.

1:42 It's thinking.

1:43 And thinking was the part I did [for] 25 years.

1:50 Building software felt like playing a video game again.

1:54 I was back.

1:56 I built 44 projects in just a few months.

2:02 And the latest one was a WhatsApp bot.

2:07 I put it on my computer, it talks through the apps that you already know.

2:17 And then I took it on a trip to Marrakesh.

2:21 You know, just to navigate around, to find restaurants, do translations.

2:28 And at first, it didn't really feel right.

2:32 It felt too much like a tool, not like a friend.

2:39 You know, too many bullet points, too many tables.

2:43 So I told it, because those modern models, they are so smart,

2:48 they know what WhatsApp is, they know how people talk.

2:51 I just had to tell it.

2:53 And then it felt right.

2:57 And you know how you talk to friends— at some point

3:00 I was walking around and I was sending it a voice message.

3:05 And then I froze because I actually, I didn't build voice in there.

3:13 I added support for images, yeah, but even that took hours.

3:16 So I was looking at the typing indicator, and then the agent responded.

3:23 And I very vividly remember this situation where

3:26 I was standing there and I was like, how did you do that?

3:30 And the agent replied, I'm not kidding you,

3:32 "The mad lad figured it out on its own."

3:36 (Laughter) And then it walked me through every single step:

3:39 how it got a message from me but there was no file ending,

3:43 so it inspected the file,

3:44 it found that it was audio, but a weird format, so it converted it.

3:48 And then it was looking for something to translate the audio,

3:52 but I didn't have it installed.

3:54 But then it found an OpenAI key, it sent the whole thing to the server,

3:58 it got it back, and then it replied.

4:01 All of that in nine seconds.

4:04 Can you imagine?

4:05 I didn't build any of that.

4:07 For me, this was the moment where I thought, this is something new.

4:12 This is not a chatbot.

4:14 Chatbots give up.

4:16 Agents improvise.

4:20 And, you know, I was sold.

4:22 I wanted to share this.

4:23 I wanted to like, tell people on Twitter.

4:26 And nobody really got it.

4:28 It's like, it's almost like you have to experience an agent.

4:33 It's kind of hard to explain.

4:37 It took me a few weeks.

4:39 And then I did something stupid.

4:42 And remember, this agent, by default,

4:47 can do anything that you can do on your computer.

4:52 So obviously I put it in a public Discord, and I invited random people.

4:56 (Laughter) And I was looking at it the whole night.

5:00 People were talking with it,

5:02 people were having fun with it, people tried to hack it.

5:06 And when my eyes were like, falling off almost,

5:12 I exited the process, and I went to bed.

5:19 Though I forgot— I built a system to be resilient.

5:23 So while I was walking to the bedroom,

5:25 the agent happily booted up again and talked to everyone in the world.

5:31 The next morning I woke up— over 800 messages.

5:36 I panicked.

5:37 I pulled the plug, I read every single message

5:42 just to see if the agent leaked my private life.

5:45 (Laughter) Nothing happened.

5:48 But it could have.

5:50 But that was the moment where it went viral.

5:54 Today, the project is called OpenClaw.

5:58 It's pretty much the fastest-growing open-source project.

6:03 Its mascot is a lobster.

6:05 It claws into your machine.

6:09 Jensen Huang calls it the operating system for personal AI.

6:15 But by far, my favorite quote is

6:17 from a friend who looked at this statistic and said,

6:20 "Peter, this is not hockey-stick growth.

6:23 This is stripper pole." (Laughter) And I, I was not ready.

6:38 You know, when something blows up like this, everything explodes.

6:43 Hundreds of messages,

6:44 reporters calling me in the middle of the night, security vulnerabilities.

6:50 And then the AI company whose model most

6:55 of my users loved sent me a trademark claim.

6:59 I had to rename the whole thing while it was taking off.

7:04 You know, they even tried to push me away from the lobster.

7:09 (Laughter) Like, I was staring at the message.

7:13 It was like, it's not even the same animal.

7:18 And then they also cut off the model most of my users loved.

7:23 You know, first the name, then the lobster, then the model.

7:32 I was that close to just deleting the whole thing.

7:36 But then I learned what people are building with it.

7:42 So at ClawCon in Vienna, because yes,

7:45 we have conferences already— (Laughter) And people wore lobster headbands,

7:52 I met Stefan and his 60-year-old dad, Gerhard,

7:58 a beer sommelier who never wrote a single line of code.

8:04 They connected OpenClaw via Bluetooth, sent it one prompt,

8:09 and the agent did the whole 90-minute brew:

8:13 temperature ramps, hop additions, everything.

8:16 And then they were like, what are we doing with all this beer?

8:22 And the agent was, "Let's make a website." So they built a website,

8:26 and then they added payments, and now they have a real product.

8:30 And almost all of it was just done via the phone.

8:36 In China, installing OpenClaw is called raising lobsters.

8:44 (Laughter) Thousands of people were lining up at the Tencent

8:50 office in Shenzhen to get their lobster installed.

8:54 Shenzhen even gives out subsidies for people running businesses on OpenClaw.

9:01 Now if you install OpenClaw on your work machine,

9:07 at least with the default settings, you might get fired.

9:12 And then I met an entrepreneur in China who showed me a spreadsheet.

9:19 Every employee, every day, one task automated by OpenClaw.

9:25 If you miss too many days, you're fired.

9:28 So fired for using it, fired for not using it.

9:37 After Marrakesh, I thought, this is incredible.

9:42 And it's also a little bit scary.

9:45 How can we make it more scary?

9:47 (Laughter) So I added a new feature, a heartbeat.

9:53 You know, by default, the agent would only wake up when you send it a prompt.

9:58 With the heartbeat, the agent would just wake up periodically,

10:02 check your emails, check your calendar, follow up on loose ends.

10:07 My initial prompt was simply "surprise me." And yes,

10:13 that's kind of as scary as it sounds.

10:16 So no large company would ship something like that.

10:20 But I'm a random builder from Austria.

10:23 I don't have a legal department.

10:25 (Laughter) I built this sandbox, this sandcastle, for me,

10:33 and I made it open-source so other people could

10:36 play with it and other people could raise their imagination.

10:41 Imagine putting an agent into a meeting.

10:45 Not for notes, we figured that out.

10:48 A bidirectional model that can listen and hear at the same time.

10:55 Somebody mentions a statistic, a subagent can spin off and check it for you.

11:00 A decision is made, the agent can send a follow-up before the meeting even ends.

11:06 In the future, we're not just going to have one agent.

11:11 You might have your work agent, your personal Claw,

11:15 maybe one for health, maybe one for relationship.

11:19 And they all should work together in a secure way.

11:23 Because how did humanity level up?

11:26 By specializing and collaborating.

11:30 And agents are about to do the same.

11:34 Imagine a company, a small business

11:38 that has ten agents that are all specialized,

11:41 taking over various parts of the business.

11:44 We don't even have a name for what it might become,

11:47 but we are about to find out.

11:50 So I created the OpenClaw Foundation, a nonprofit, open-source forever.

11:58 Because what OpenClaw did for many people was it moved AI from this scary,

12:07 nebulous thing into something that is fun and useful and maybe a bit weird.

12:15 You know, lobsters and headbands and beer businesses.

12:20 Because what we need in the future is more people spending more time

12:26 with AI to better understand how

12:29 powerful and transformative this technology really is.

12:34 In New York, at ClawCon...

12:40 (Laughter) Yes, they're everywhere now.

12:44 There were thousands of people that were

12:47 discussing what the lobster did this week.

12:52 A retiree in Shenzhen who automated their groceries.

12:58 A teenager in Sao Paulo who built a tutoring business on OpenClaw.

13:04 Gerhard and his beer machine.

13:07 None of them are programmers.

13:10 All of them are builders.

13:13 Because that's the real transformation.

13:16 It's not a technology.

13:18 It's the access.

13:20 Agents change who can build things, and that door is not closing again.

13:27 Because when you can prompt a prototype into existence in one hour,

13:35 anything is possible.

13:37 The next breakthrough can come from anyone, any country, any cafe.

13:46 When even a burnt-out founder, staring at the screen,

13:53 wondering if his spark is gone can do something like that.

14:01 It's not gone.

14:03 It's just waiting.

14:05 The lobster is loose, and it's not going back into the tank.

14:10 Thank you very much.

14:12 (Applause) Chris Anderson: Come here.

14:24 I mean.

14:27 I think I want to say something personal to you.

14:30 With love, but truth.

14:33 You really terrify me.

14:35 (Laughter) CA: I’m serious.

14:39 If Hollywood was to ever make a movie in which humanity opened Pandora’s box,

14:45 and everything went crazy, like,

14:47 you seriously could be cast as the star character.

14:50 (Laughter) CA: Because the story we’re told is

14:54 that AI researchers are doing all this great stuff,

14:57 but they're taking all these great efforts to ensure

15:01 safety and make sure nothing bad could happen.

15:04 You take glee in seeing what might happen if you just put it out there.

15:10 Like, is any part of you feeling that that's a little bit reckless?

15:17 Peter Steinberger: I wouldn't say so.

15:19 I see my work as a...

15:20 (Laughter) PS: I see it as a window into the future.

15:24 Like, in the very beginning, there were all these scary moments.

15:31 Now we have proper security layers.

15:33 You can have your sandbox, you can put your lobster into a very small,

15:39 tiny box and really control what it can do.

15:43 There are still some issues that we need to figure out,

15:46 but the fact that so many people want

15:48 this now will help to figure this out much faster.

15:52 CA: So I'm glad you mentioned the security layers.

15:56 I mean, you're making a huge bet on human ingenuity using this.

16:01 This is an incredible tool that suddenly can

16:04 maximize the power of what any human can do.

16:12 How many people in your community are taking

16:16 the safety issue seriously and want to use OpenClaw,

16:19 for example, to find smarter ways?

16:20 Just checking whether anything might be going wrong and giving

16:25 an early alarm to someone or something like that?

16:28 PS: Most people are not as reckless [as]— number one,

16:31 putting it into a public Discord— strongly don't recommend.

16:36 Number two, I think I single-handedly

16:39 increased Mac mini sales by multiple percent.

16:43 So most people give it their own little Mac mini.

16:46 Mine's a little princess.

16:48 Mine got a Mac studio.

16:49 It calls it “The Castle.” And that greatly reduces the actual

16:54 risk because you can only access what's on that computer.

16:57 And maybe all your pictures are not there.

17:00 CA: Well, definitely, if humanity goes down,

17:02 I'll be very grateful for at least the rise

17:04 in Mac mini sales for a period of time.

17:07 You are amazing.

17:08 You are amazing, and I think you're

17:11 actually right at the cutting edge of whether

17:14 AI is going to be the biggest boon ever or possibly a serious problem.

17:19 And, you know, I hope you continue conversations

17:21 with people here and just help us get smarter on how to do this the right

17:26 way because it is absolutely incredible what you've built.

17:29 Thank you for sharing so honestly.

17:31 PS: Thank you so much.

17:32 (Applause)