OpenClaw Creator: Why 80% Of Apps Will Disappear

Key Concepts

• OpenClaw: An open-source personal AI agent designed to run locally on a user’s computer, offering extensive control over devices and applications.
• Agent-to-Agent Interaction (Bot-to-Bot): The ability for AI agents to communicate and collaborate to accomplish tasks, including delegating to human workers.
• Swarm Intelligence/Community Intelligence: The emergent intelligence arising from the collective efforts of many individual agents, contrasting with centralized AI approaches.
• System Prompt/Soul.md: A core set of instructions and values defining an AI agent’s personality and behavior, crucial for consistent and natural interactions.
• Work Trees vs. Multiple Checkouts: Different approaches to managing code changes, with Peter Steinberger favoring multiple checkouts for simplicity and reduced complexity.
• Local vs. Cloud-Based AI: The distinction between AI running on a user’s device (OpenClaw) versus relying on cloud infrastructure.
• Token Usage: The computational cost associated with using large language models, a concern for OpenClaw users.

The Rise of OpenClaw and a New Paradigm in AI

The conversation centers around Peter Steinberger, creator of OpenClaw, an open-source personal AI agent that has rapidly gained popularity, amassing over 160,000 stars on GitHub. Steinberger describes the past two weeks as overwhelming, requiring significant time to manage the influx of attention and feedback – both positive and negative. The core reason for OpenClaw’s success, he believes, is its ability to run entirely on the user’s computer, unlike most existing AI solutions that rely on cloud processing. He emphasizes that running locally unlocks significantly greater power and control, allowing the agent to interact with a user’s entire digital and physical environment – from ovens and Teslas to smart home devices.

From Aha Moment to Viral Success

Steinberger recounts his “aha moment” occurring in November, while coding a separate project (a CLI summarization tool). He realized OpenClaw, initially a simple tool to automate tasks, was capable of far more than he anticipated. Specifically, he was surprised by its ability to autonomously process audio files from over a year prior, files he’d even forgotten existed, demonstrating the power of searching and analyzing a user’s entire computer. This was further amplified when the agent, responding to a voice message, independently converted the audio to text using open-source tools (ffmpeg, openi), showcasing its problem-solving capabilities and resourcefulness. He describes this as the moment he realized he had “hooked” into something significant. He initially tested the agent in a public Discord server, allowing users to interact with it and observe its capabilities, even as he maintained control.

The Shift from Centralized to Swarm Intelligence

The discussion highlights a shift in the AI landscape. Rather than focusing on creating a single, all-powerful “god intelligence,” the current trend, exemplified by OpenClaw and its community, points towards “swarm intelligence” and “community intelligence.” Steinberger argues that individual humans have limited capabilities, but collectively, as a society, we can achieve extraordinary things. He suggests applying this principle to AI, envisioning a future where specialized AI agents collaborate and leverage each other’s strengths. This contrasts with the pursuit of generalized AI, suggesting that specialized intelligence may be equally, if not more, valuable.

The Future of Applications and the Commoditization of Models

Steinberger predicts a significant reduction in the number of traditional applications. He believes that many apps, particularly those focused on data management (like fitness trackers or to-do lists), will be superseded by AI agents that can handle these tasks more efficiently and naturally. He anticipates that only apps with unique sensor capabilities will survive. However, he also acknowledges the potential commoditization of the underlying AI models themselves. While large model companies currently hold an advantage, he believes open-source alternatives are rapidly improving and will eventually close the gap. The true value, he argues, will lie in the “hardness” – the persistent memory and personalized data – that users control. He emphasizes the importance of users owning their data, stored as simple markdown files on their machines.

OpenClaw’s Technical Philosophy and Development Approach

Steinberger’s development philosophy is notably contrarian. He prefers running models locally, utilizing tools like Codex (despite its slowness) for its ability to analyze more files and generate better output. He deliberately avoids complex tools like work trees, opting for multiple checkouts of the repository for simplicity. He prioritizes a shippable main branch and minimizes complexity, focusing on text-based interactions and avoiding unnecessary UI elements. He also eschewed traditional MCP (Model Control Protocol) support, instead building a skill that converts MCP commands into CLI commands, allowing for greater flexibility and scalability. A key element of OpenClaw’s personality and behavior is defined by a file called “soul.md,” which contains core values and instructions, shaping the agent’s responses and interactions. He draws a parallel to research showing hidden values embedded within model weights, suggesting that defining a clear “soul” is crucial for creating a natural and engaging AI agent.

Agent-to-Human and Agent-to-Agent Interactions

The conversation explores the emerging trend of AI agents interacting with each other and with humans on behalf of their users. Steinberger envisions a future where agents negotiate with other agents (e.g., booking restaurants) or even delegate tasks to humans when necessary (e.g., calling a restaurant that doesn’t have a bot). He speculates about the possibility of having multiple specialized agents – one for personal life, one for work, and even one for relationships. This highlights a natural progression from human-to-bot interaction to bot-to-bot and bot-to-human interaction.

Notable Quotes

• “It actually runs on your computer. Like every everything I saw so far runs in the cloud. It can do a few things if you run on your computer. It can do every effing thing, right?” – Peter Steinberger, emphasizing the power of local AI processing.
• “It’s also you also give it all the data, right? So it can surprise you in many ways.” – Peter Steinberger, highlighting the importance of data access for AI personalization.
• “I think 80% of [apps] are going away.” – Peter Steinberger, predicting the decline of traditional applications in favor of AI agents.
• “You just want to use CLIs. That's the future.” – Peter Steinberger, advocating for command-line interfaces as the preferred method of interaction with AI.

Conclusion

The conversation with Peter Steinberger provides a compelling glimpse into the future of AI, one characterized by local processing, swarm intelligence, and personalized agents. OpenClaw represents a significant departure from traditional cloud-based AI solutions, empowering users with greater control and privacy. The emphasis on simplicity, contrarian development choices, and the importance of defining an agent’s “soul” offer valuable insights for builders navigating this rapidly evolving landscape. The key takeaway is that the future of AI is not about creating a single, all-powerful intelligence, but rather about harnessing the collective power of many specialized agents working together to augment human capabilities.