OpenClaw Founder's Firsthand Account: How AI Agents Are Reshaping the Future of Software Engineering

The viral success of OpenClaw, an open-source AI assistant, stands as one of the most representative cases in the AI application space in recent years. This free, locally-run AI assistant that interacts with users through everyday chat apps catapulted its founder, Peter Steinberger, into the industry spotlight in record time — ultimately leading him to join OpenAI, where he now helps shape the future of personal AI agents.

In an in-depth interview, Peter candidly shared his understanding of agentic coding, the real cost behind going viral, and a counterintuitive core insight: in an era where AI capabilities are advancing at breakneck speed, the "human touch" has paradoxically become more precious than ever.

From Writing Code to "Commanding an Agent Army": A Fundamental Shift in the Engineer's Role

Peter's definition of software engineering is undergoing a fundamental transformation. He states bluntly: "We've reached the point where we don't have to write the code ourselves." But this doesn't mean the engineer's job has gotten easier — rather, the center of gravity has completely shifted.

"What matters is thinking: what can I do to help my agents do the best work?" Peter calls this process designing a pipeline from "what I want" to "here's the final result." The engineer's new role is to design the loops or graphs that agents can enter to test and review code, and to optimize where their own attention should be directed.

He cuts straight to the point that these so-called loops and graphs are nothing mysterious: "What have engineers been doing for the past 50 years? We write automation scripts, build small tools. Now we just have AI as an additional tool that lets us automate more and do more."

Mindset Shift: Managing AI Agents Like Managing a Team

Peter believes the key to working efficiently in the AI era comes down to mindset. He shares a profound insight: why do some people see massive productivity gains with AI while others find it "completely useless"?

The answer comes from his experience running a startup and managing teams. When someone transitions from "I write the code" to "I direct a group of people to write code for me," they must accept a reality — the code won't be exactly as they would have written it; there will be slight differences. If you micromanage real human reports, they'll quit. Agents won't quit, but the results are equally bad.

"Maybe it's not exactly what I wanted, but maybe that doesn't matter anymore. Can it get me to my goal? Yes." This willingness to let go is what Peter sees as the dividing line between "AI beneficiaries" and "AI skeptics." Many of his open-source projects are handled automatically by an open-source orchestrator: when a new issue comes in, it automatically builds a PR and runs tests, and he only steps in as an "executive" when decisions are needed.

Counterintuitive Advice: Keep Your Typos, Stay Authentically Human

The most thought-provoking point in the interview was Peter's emphasis on the value of the "human touch." In a world where AI can easily generate massive amounts of content, he actually advises people to "keep your typos, maintain that raw feel."

"If you write something for your website and then use AI to make it more professional, it'll add those telltale signs. People can no longer tell whether your blog post was carefully thought out and lightly polished, or whether it was two pages generated from a three-word prompt. Once they catch that scent, they'll consciously or subconsciously devalue it."

He admits he receives a flood of cold outreach emails every day, and the moment he spots AI writing markers like EM dashes, he immediately blocks the sender. "If the email reads like AI slop, that'll be the only reply you ever get."

This perspective extends into a profound question for our era: when capturing attention (eyeballs) becomes the scarcest resource, the most effective differentiation strategy turns out to be "just being a real person." Peter even says he'd rather receive a PR with typos and a handwritten feel, or one that includes operation logs as "proof of effort."

The Art of Saying "No": Systems Thinking in the AI Era

Peter candidly admits one mistake he made with OpenClaw was "saying yes to too many things." In the AI era, conjuring up a new feature out of thin air has become incredibly easy, but the feature itself was never the problem.

"The problem is those other 99 features that should interact with the new one, and thinking about how it fits into your entire system." Peter points out this is precisely where agents fall short — every time an agent enters a codebase, it "knows nothing, navigates like a horse" finding its way from scratch. They can navigate large codebases well, but they lack holistic understanding of the system and can't judge whether "this actually makes sense" or "does this contradict something over there?"

"That's exactly where you excel." He uses an analogy from a comic he read as a child: a chef makes a cake, and everyone wants to add one ingredient. Each one looks fine on its own, but add 50 at once and you've got a terrible cake. He brought in too many well-intentioned maintainers with great ideas, and that's exactly what happened.

The Price of Going Viral: A "Blitzkrieg" That Nearly Broke the Founder

OpenClaw's viral success wasn't the result of carefully orchestrated marketing, but Peter is candid: "It wasn't organic, but I also wish it hadn't been so intense — it nearly broke me."

His product instinct came from repeated "emotion tests": when he posted on Twitter, nobody understood, but whenever he showed the product to friends, it always triggered strong emotional reactions — some found it magical and fun, others felt scared or creeped out. "That's the key to building products: create something that makes people feel something strongly."

The real turning point came in early January: he set up a Discord server, put the agent in it, publicly demonstrated how he collaborated with it, and let users interact directly. This move made the previously one-on-one "wow effect" scalable. He clearly observed the path of viral spread — from an influential KOL, to a bilingual fan in Japan, penetrating social circles layer by layer, until one maintainer exclaimed: "Oh my God, it's reaching the normies!"

With virality came chaos: his identity was leaked, phone calls around the clock, a deluge of emails, and project attention demands that surged a hundredfold. "If you want to go viral, we prepared for going viral, but things get very, very hard."

The Open-Source Product Dilemma: Caught in the Security Controversy Crossfire

What hit Peter hardest emotionally was the avalanche of security blog posts and security reports. "So much exaggeration, fabricated claims, everyone optimizing for clicks." He acknowledges there were real issues behind it — giving agents more capabilities does expand the attack surface.

Driven by a sudden sense of responsibility, he spent months doing almost nothing but one thing: making the product secure. But he reflects: "It really hurt the product. At some point, I stopped building for myself and started building for others, adding a lot of things I didn't even agree with."

He also criticizes certain aspects of security culture: some people would privately send him "five things you must fix immediately" — architecture-level challenges — while demanding all kinds of press, CVEs, and coordination processes, making everything incredibly complicated. "But this is an open-source project — why not just submit a PR?"

This experience led Peter to distill two iron rules of product building: First, you must use it yourself, otherwise you're building the wrong thing. Second, you should enjoy it. "This thing was incredibly draining, I wasn't enjoying it at all, I lost the fun, and it showed in the product."

Open Source, Closed Source, and the Future of Software Engineering

On the open-source versus closed-source debate, Peter holds a view so pragmatic it's almost subversive: "It doesn't matter anymore. Even if you package it as an Electron app, it's essentially still open source — harder for humans to read, but not much harder for agents."

He believes the value of keeping code closed is shrinking because "I can always replicate it." The real value lies in "who's behind it, who's updating it. Software that isn't updated is dead, and everything is changing faster and faster." He personally despises software that can't be modified through prompts — this is exactly what motivates him to build so many of his own tools: when something breaks, instead of going through customer support, he just has the agent fix it directly.

Looking ahead, Peter says what fascinates him even more than OpenClaw itself is "how to build software with agents in the future." He's pursuing a goal: using OpenClaw to build OpenClaw itself. He's also observed an interesting phenomenon — many features he experimented with early on (like having an agent automatically check tasks every five minutes) have now become the default expectation for everyone.

Regarding business model, retention, defensibility, and other questions investors commonly ask, Peter admits OpenClaw should be doing more in this area, such as adding telemetry to understand which features can be removed — "otherwise I'm flying blind, working on vibes." But he also emphasizes that with a true open-source project, "you're building it for yourself; you don't need massive user penetration targets."

Conclusion: Being Authentically Human in the AI Era

Peter Steinberger's story encapsulates the core tension of today's AI application wave: technological capabilities are unprecedentedly powerful, yet tools haven't caught up with this "ability overhang"; agents can replace vast amounts of coding work, but they can't replace systems thinking and aesthetic judgment; AI can generate infinite content, yet human authenticity and warmth have become the scarcest moat.

His advice is simple yet powerful — invest in your personal brand, stay authentic, enjoy what you do, and learn to say "no." In an era where anyone can mass-produce with AI, perhaps the greatest competitive advantage lies precisely in those things that machines cannot fake.