Peter Steinberger, creator of the viral AI agent OpenClaw and subsequently hired by OpenAI, has advice for those experimenting with AI technologies, including AI agents. From his own experience, the best way to build today is to explore, be playful, and not expect to become an expert at what you do right away.
“I’d like to say there was a unified plan from the beginning, but a lot of it was just exploration,” Steinberger said. “I wanted things, but they didn’t exist. And… let’s say I inspired them to exist.”
Developers were chatting with Romain Huet, Head of Developer Experience at OpenAI, on the first episode of the company’s new Builders Unscripted podcast. Here he talked about what OpenClaw was like in its early days and how there was no plan when it started.
Steinberger explained that he started by building a tool to integrate with WhatsApp, but he expected the AI Institute to build a tool similar to what he was working on in the near future, so he put that aside a bit and focused on other things.
“I just experimented a lot. My mission was kind of to have fun and inspire people,” Steinberger said. But as of last November, developers were surprised to find that not a single AI lab had started building what they wanted to use. This led him to create the first prototype of what is now OpenClaw.
“What really clicked for me was this weekend trip to Marrakech, and I found myself using it more and more because it was so convenient. We didn’t have really good internet,[but]WhatsApp works everywhere,” he said. This tool has made it easy to find restaurants, do research on your computer, and send texts to friends.
The more Steinberger played with technology, the more he realized how good modern AI models were at problem-solving, just like programmers.
“Now they can actually come up with solutions on their own, even if they’ve never programmed something before,” he said.
Steinberger says the workflow has improved throughout the build process and emphasizes to other developers not to give up as this will take time.
“There are people writing software the old-fashioned way, and the old way will become obsolete,” he pointed out. They then decided to try vibe coding, but were disappointed with the results.
“I think vibecoding is a slur,” Steinberger says, essentially suggesting that it’s not as simple a process as words make it sound at first. “They’re experimenting with AI, but they don’t understand that it’s a skill,” he said, likening the process of coding with AI to learning to play the guitar.
“You’re not going to be good at guitar from day one,” he said. Instead, we encourage people to approach learning with a more playful attitude. If you write a prompt now, you intuitively know how long it will take, and if it takes longer, reflect on what went wrong and adapt.
“My advice is always to approach it in a playful way. Build something you’ve always wanted to build. If you’re at least a little bit of an architect, you probably have something in the back of your mind that you’ve always wanted to build. Just play around with it.”
This ability to experiment and have fun is paramount, especially now that people are worried that their jobs will be taken over by AI.
“If your identity is: I want to create things. I want to solve problems. If you have a lot of initiative and you’re smart, you’re going to be more in demand than ever,” Steinberger said.
