Peter Steinberger, known for creating the hugely popular AI agent OpenClaw and now part of OpenAI, offers insights for individuals venturing into AI technology, particularly AI agents. Based on his journey, he recommends that the most effective approach to building currently involves exploration, a playful attitude, and understanding that expertise isn’t immediate.
“While I’d love to claim I started with a clear, overarching plan, much of it truly involved discovery,” Steinberger shared. “I had certain desires, these things weren’t available, so … I essentially prompted them into being.”
Steinberger recently discussed his early experiences with OpenClaw and his lack of initial planning during a conversation with Romain Huet, OpenAI’s Head of Developer Experience, on the inaugural episode of OpenAI’s new podcast, Builders Unscripted.
He elaborated that his initial project was a WhatsApp-integrated tool, which he temporarily paused to pursue other interests, anticipating that major AI labs would soon develop similar functionalities.
“I simply engaged in extensive experimentation. My objective was primarily to enjoy myself and motivate others,” Steinberger mentioned. Yet, by November of the previous year, he was astonished to find that no AI research facilities had begun developing the kind of tool he envisioned. This realization prompted him to create the foundational prototype that evolved into OpenClaw.
“The real breakthrough happened during a weekend trip to Marrakesh, where I noticed myself relying on it much more due to its sheer convenience … Internet access was poor, but WhatsApp functions universally,” he recalled. This tool greatly simplified tasks like locating restaurants, researching information on his computer, and messaging friends.
As he continued to interact with the technology, Steinberger observed the impressive problem-solving capabilities of contemporary AI models, akin to those of human programmers.
“These models can now autonomously generate solutions, even without any explicit programming from your side,” he pointed out.
Steinberger shared that his development process progressively refined, emphasizing to fellow developers that such improvements require time and perseverance.
“Some individuals continue to … develop software using traditional methods, which are becoming obsolete,” he noted. When these developers then attempt ‘vibe coding,’ they often find the outcomes unsatisfying.
“I consider ‘vibe coding’ to be a misleading term,” Steinberger remarked, implying that the initial process is more complex than the phrase suggests. “People experiment with AI without realizing it’s a learned skill,” he explained, drawing an analogy to mastering a musical instrument like the guitar.
“One doesn’t achieve guitar proficiency on the very first day,” he stated. He advocates for a more recreational approach to learning. Currently, when crafting a prompt, he instinctively gauges the required time; if it takes longer, he analyzes the potential issues and adjusts his strategy.
“My consistent counsel is to engage with it playfully. Create something you’ve always wished to build. If you possess even a slight inclination towards creation, there’s likely a project you secretly aspire to make. Just immerse yourself in it.”
The capacity for experimentation and enjoyment holds paramount importance, particularly in an era where concerns about AI replacing human jobs are prevalent.
“If your core identity revolves around creating and problem-solving, if you are proactive and intelligent, your skills will be more sought after than ever before,” Steinberger affirmed.