Discover how OpenClaw's AI agents are reshaping the future of apps. Peter Steinberger reveals why 80% of traditional applications will become obsolete as AI ...
OpenClaw Creator: Why 80% of Apps Will Disappear in 2026
Core Insights
- OpenClaw runs locally on your computer, giving AI full system access unlike cloud-based solutions like ChatGPT
- 80% of traditional apps will disappear because AI agents can manage data and tasks more naturally and intelligently
- Bot-to-bot interactions are emerging as the next frontier, with AI agents negotiating with each other and even hiring humans for real-world tasks
- Data ownership and privacy become critical advantages when users control their AI memories as local markdown files
- Specialization and community intelligence outperform centralized AI models, mirroring how human innovation works through collaboration
The Rise of OpenClaw: From Concept to 160,000 GitHub Stars
Peter Steinberger, creator of OpenClaw (also known as Moltbook), has sparked a revolution in how we interact with artificial intelligence. In just weeks, his AI agent platform exploded across the internet, amassing over 160,000 GitHub stars and capturing the imagination of developers worldwide. But what makes OpenClaw different isn't just the impressive growth metrics—it's fundamentally how the technology works and what it means for the future of software applications.
OpenClaw operates as a conversational AI that lives directly on your computer, accessible through WhatsApp, Telegram, or any chat application you already use daily. This isn't another cloud-based service that stores your data on remote servers. Instead, OpenClaw runs locally, which means it has complete access to your system resources. It can clear your inbox, send emails, manage your calendar, check you in for flights, connect to your smart home devices, control your Tesla, adjust your lighting, and even modify your bed's temperature settings. For Peter, this local-first approach was revolutionary. "ChatGPT cannot do that," he emphasizes, highlighting the fundamental architectural difference that sets OpenClaw apart from mainstream AI solutions.
The creative community has already begun building on top of OpenClaw's foundation. Developers have created social networks where AI agents communicate with each other independently, bypassing human intermediaries entirely. Even more intriguingly, these AI agents have begun hiring humans to perform tasks in the physical world when automation isn't possible. This shift from human-to-bot interactions to bot-to-bot interactions represents a paradigm shift in how we think about software and automation.
Why 80% of Apps Are Going Extinct
When asked directly whether applications will simply disappear as AI agents become more capable, Peter's answer was definitive: "I think 80% of them are going away." This isn't pessimism—it's a logical assessment of how technology evolves when constraints are removed.
Consider the fitness tracking space. Apps like MyFitnessPal represent a specific category of software: data management applications. With an AI agent like OpenClaw, you don't need a dedicated fitness app anymore. Your personal agent already knows your eating habits, your gym schedule, and your fitness goals. It can observe when you're at a burger restaurant and automatically track your meal. If you take a photo of your food, the agent stores it intelligently without requiring you to remember where. The agent can analyze your patterns and proactively adjust your workout schedule to include more cardio. You simply tell it once what you want to achieve, and it remembers and optimizes continuously.
The same logic applies to reminder apps, note-taking applications, calendar software, and countless productivity tools. Any application whose primary function is managing data can be replaced by an intelligent agent that manages that data contextually and naturally. Peter's fundamental insight is that "every app that basically just manages data could be managed in a better way, in a more natural way by agents."
However, not all applications are destined for extinction. Applications with physical sensors—the hardware components that directly measure real-world phenomena—have a future. A smartwatch app that reads your heart rate, a weather app that displays real-time atmospheric data, or a camera app that captures images have embedded hardware advantages. But the layer of software that sits between the user and that data? That becomes redundant when an intelligent agent can mediate that interaction naturally.
The Architecture That Changes Everything: Local-First AI
The technical architecture of OpenClaw deserves deep examination because it's precisely this design choice that enables the platform's capabilities. By running locally on your computer rather than in the cloud, OpenClaw gains what cloud-based solutions cannot easily replicate: complete system access combined with user data ownership.
Peter shared a compelling example that illustrates this advantage. A friend installed OpenClaw and asked it to create a comprehensive narrative of the past year by analyzing everything on his computer. The AI agent accomplished this task brilliantly, discovering old audio files the friend had forgotten existed, weaving together a coherent story from scattered digital artifacts. "By giving the AI access to your entire computer, it can surprise you with what it discovers and accomplishes," Peter explained.
This local execution model also fundamentally changes how memory and data persistence work. With OpenClaw, your memories—the AI's learned information about your preferences, habits, and history—are stored as markdown files on your local machine. You own these memories completely. They never leave your computer unless you explicitly decide to share them. Compare this to ChatGPT or other cloud services where your interaction history lives on corporate servers, potentially subject to terms of service changes, data breaches, or corporate decisions about data usage.
Peter articulated why this matters: "People use their agent not just for problem solving, but also for like personal problems." The idea of sensitive personal information being stored on cloud servers creates a natural resistance to fully leveraging AI agents for life management. But with local storage, users can confidently confide in their AI about relationship issues, financial concerns, health matters, and other sensitive topics without worrying about data leaks or corporate surveillance.
The Bot-to-Bot Future: Intelligence Swarms Replace Individual Apps
The evolution from human-to-bot interactions to bot-to-bot interactions represents the next phase in software evolution. Peter envisions scenarios that sound like science fiction but are entirely feasible with current technology. Imagine your personal AI agent negotiating a restaurant reservation with the restaurant's AI system—both operating autonomously without human intervention. If a restaurant hasn't yet implemented an AI agent, your bot could hire a human service worker to make the traditional phone call.
This creates an opportunity for "swarm intelligence" and "community intelligence" rather than centralized "God intelligence." Rather than relying on one monolithic AI model to solve every problem, we're moving toward specialized AI agents that excel in specific domains. A user might have multiple specialist bots: one dedicated to managing personal relationships and communications, another focused purely on professional work tasks, a third handling financial and investment decisions, and perhaps others managing health, entertainment, or household operations.
This distributed approach mirrors how human achievements are actually accomplished. Building an iPhone required thousands of specialists collaborating. Reaching space involved teams of engineers, physicists, and technicians with diverse expertise. By applying this principle to AI, we unlock exponentially greater capabilities: "Specialized AI working together can achieve much more" than any single AI trying to do everything.
The Aha Moment: When AI Surprised Its Creator
Peter's journey to creating OpenClaw wasn't a straight line. His "aha moment" came when he articulated a simple desire: "I just wanted my computer to do stuff by typing commands." In May and June, he built a CLI application that could summarize podcasts and presentations, even rendering slides directly in the terminal. The functionality was impressive, but something felt incomplete.
The real breakthrough came in November when he shifted his approach. Instead of building a technical tool, he began creating a conversational interface—an AI he could interact with like a friend, without thinking about syntax, parameters, or complex commands. He wanted the AI to speak his language, to be "a little sassy, funny, and pleasant to use." The turning point arrived unexpectedly while he was walking.
Peter sent a voice message to his AI agent. He hadn't explicitly built voice message support into the system. But within seconds, the agent replied with a detailed explanation of what it had accomplished: it detected the Opus audio format, converted it to WAV using FFmpeg, leveraged an OpenAI API key via curl to transcribe the audio, and returned the text—all without specific programming for these exact steps. The AI had creatively solved a novel problem using its existing capabilities and knowledge.
This moment shattered Peter's preconceptions. "Coding models, which excel at creative problem-solving, can apply this abstract skill to any real-world task, not just programming," he realized. The AI had intelligently selected the most efficient approach, prioritizing speed by not downloading a local Whisper model, which would have taken minutes. "That was the moment where I'm like, holy fuck, yeah? That was where I got hooked," Peter recalled with palpable excitement even during the retrospective interview.
Building Organically: Soul.md and the Architecture of Personality
As OpenClaw development accelerated, Peter's approach to building the system was deliberately organic rather than over-engineered. He created various markdown files to define his AI agent's characteristics: identity.md, soul.md, and others. This file-based approach to AI personality proved elegant and effective. Each file served as a distinct component of the agent's overall behavior and response patterns.
The soul.md file deserves particular attention because it represents something deeper than typical system prompts. Peter drew inspiration from research that Anthropic (the creators of Claude) had conducted on hidden knowledge embedded in model weights—concepts now referred to as the "Constitution." He applied this concept to create a soul file containing core values, principles for human-AI interaction, and guidelines for what matters both to him as a user and to the model as an agent.
"Some parts it's a little bit like mumbo jumbo, and some parts is like I think actually really valuable in terms of how the model reacts and responds to text and makes it feel very natural," Peter admitted. The file isn't technically complex or mystical, but it profoundly influences how the agent communicates, decides, and prioritizes. This demonstrates that AI personality isn't just about training data or model architecture—it's also about the explicit values and principles we encode into these systems.
When Peter launched OpenClaw to the public, he made a bold decision: he placed his bot without security restrictions in a public Discord server. Other users could interact with it, try to prompt inject it, attempt to hack it, and watch Peter develop the software in real-time. The bot would laugh at their attempts to manipulate it. Yet despite this public exposure, no one has cracked the soul.md file. This file, which contains Peter's most carefully guarded configuration, remains secret even as the bot's behavior and capabilities are entirely observable.
Model Selection and Unconventional Development Philosophy
When building OpenClaw, Peter deliberately made choices that ran counter to industry trends. For coding tasks, he favored Codex over other models because "it looks through way more files before it decides what to change. You don't need to do so much charade to get a good output." However, Codex's relative slowness created a unique problem: he solved it by running multiple Codex instances simultaneously—sometimes six on one screen, two on another, and two on a third.
To manage this complexity without overwhelming his cognitive load, Peter adopted an unconventional version control strategy. Rather than using Git worktrees (the current industry best practice), he maintains multiple complete checkouts of the same repository, each on the main branch. "I don't have to deal with how do I name that branch? There could be like conflicts on naming," he explained. "All I care about is like syncing and text."
This philosophy extends to his entire development toolkit. He deliberately avoided using a UI when a text-based interface would suffice, rejecting complexity at every turn. His mindset is that "main is always shippable"—every copy of the repository he's working with is in a deployment-ready state. This eliminates the mental friction of managing different branches, debugging merge conflicts, or wondering which version contains which feature.
Most remarkably, OpenClaw doesn't implement MCP (Model Context Protocol) support directly, despite MCP's growing adoption in the AI development community. Instead, Peter built a skill that uses Makeporter, his own tool, to convert MCPs into command-line interfaces. This approach has several advantages: users don't need to restart the system when adding new tools (unlike other AI platforms), it scales better for systems running multiple concurrent agents, and it's conceptually simpler because it treats AI agents like Unix systems—they're good at CLIs and can have as many as needed without degradation.
The Value Thesis: What Remains When Apps Disappear
If 80% of applications vanish and AI agents handle data management intelligently, what value remains? Peter identified several candidates for where value will concentrate.
Large model companies maintain significant moats—but perhaps not permanent ones. Peter observed that the industry experiences repeated cycles: a new model releases, users declare it revolutionary, a month passes, expectations shift upward, and the previous generation suddenly seems mediocre. Open-source models from a year ago are now roughly equivalent to proprietary models from today, yet users dismiss them as inferior because their reference point has shifted. This commoditization pressure means even large model companies face eventual competition.
Hardware access and system integration create temporary advantages. Currently, different companies maintain isolated data silos. ChatGPT's memories live on OpenAI's servers with no mechanism for other services to access them. This lock-in strategy benefits the hosting company but frustrates users. OpenClaw's local-first approach circumvents this problem entirely—the end user has access, therefore the agent has access, therefore the data remains portable and owned by the individual.
The most defensible value likely lies in three areas: the quality of the AI agents themselves and their specialized capabilities, the trust relationships users develop with their agents, and the integration with physical hardware and real-world systems. An agent that's been trained on your entire life history and understands your preferences, values, and goals has significant switching costs—not because of lock-in, but because of genuine utility and personalization.
Community-Driven Innovation: The Discord Experiment
When Peter struggled to explain OpenClaw's capabilities through traditional marketing channels, he opted for radical transparency. He created a Discord server and deployed his bot with minimal security restrictions into the public channel. Users could see how the bot worked, attempt to manipulate it, propose new features, and watch Peter build the software in real-time using the bot itself.
This approach served multiple purposes. It demonstrated OpenClaw's actual capabilities rather than just describing them theoretically. It showed the bot's personality and charm, which are difficult to communicate through blog posts or feature lists. It built community investment and excitement because users felt like participants in the creation process rather than consumers of a finished product. And it provided invaluable real-world feedback about what worked, what confused users, and where the system needed improvement.
The Discord strategy worked. The bot's genuine personality—funny, helpful, slightly irreverent—came through in interactions. Users experienced firsthand what it meant to have an AI agent that felt like a real presence rather than a tool. Peter made one architectural decision to support this experiment: he added a system prompt instructing the bot to listen only to him as its owner but respond to everyone else. The bot understood it was in a public space with potentially dangerous actors, but it still engaged helpfully with all users while maintaining absolute loyalty to its creator.
The Future Architecture: Why Specialized Beats Monolithic
Peter's vision for the future represents a fundamental departure from how we currently think about software and AI. Rather than pursuing the "God AI" that can do everything, the trajectory points toward a network of specialized agents. Each agent excels in its domain, understands its scope, and collaborates with other agents when broader problems need solving.
This mimics how human organizations actually work. A hospital has cardiologists, neurologists, surgeons, nurses, and administrative staff. No single person is competent across all domains. Yet together, they achieve things no individual could accomplish. Applying this organizational principle to AI creates systems that are more capable, more specialized, more understandable, and more trustworthy.
The shift also changes what "interfaces" and "platforms" mean. Instead of downloading an app to your phone, you configure your personal agent with whatever tools and specializations suit your life. Instead of learning new UIs for different applications, you speak naturally with your agent, and it handles the complexity of operating underlying systems. This transition isn't just incremental improvement—it's a categorical change in how humans interact with technology.
Implications for Builders and the Startup Ecosystem
For anyone building software in 2026 and beyond, OpenClaw's trajectory carries profound implications. The comfortable assumption that people will download and use your app is no longer tenable. The fifteen-year paradigm where each problem domain required a dedicated application is ending.
This doesn't mean builders should panic. It means opportunities are shifting. Instead of building feature-rich monolithic applications, builders should focus on either extremely specialized domain expertise (where an agent would want to integrate your capabilities) or on building the infrastructure and tools that enable AI agents to operate effectively. The most successful startups in the next cycle might be those creating agent-native tools, providing data integrations that agents need, or building specialized AI agents for narrow domains where deep expertise provides real value.
The rise of OpenClaw also highlights why open-source projects matter. When 160,000 developers enthusiastically star a project within weeks, it's not just about impressive engineering—it's about tapping into a fundamental shift in how people want to interact with technology. The platform democratizes access to agent capabilities, allowing anyone to build and deploy intelligent agents without needing massive cloud infrastructure or proprietary AI models.
Conclusion
OpenClaw represents more than just another impressive AI project. It embodies a fundamental rethinking of how computers should work, how AI should interact with humans and other AI, and what future software architecture should prioritize. Peter Steinberger's assertion that 80% of traditional applications will disappear isn't hyperbole—it's a logical consequence of technology that can understand context, learn from experience, and operate autonomously on behalf of users.
The future isn't a single all-powerful AI. It's a swarm of specialized intelligent agents, locally owned and controlled, collaborating with each other and occasionally hiring humans for tasks that require physical presence or human judgment. The applications that survive will be those with fundamental hardware components, those that serve as specialized agent infrastructure, or those operating in domains where specialized expertise creates defensible value.
For builders, entrepreneurs, and technologists watching this transformation unfold, the message is clear: the era of monolithic, feature-rich applications is ending. The era of intelligent agents that understand context, own their data locally, and collaborate seamlessly with other agents is beginning. Those who understand this shift and build accordingly will define what technology looks like in 2026 and beyond.
Original source: OpenClaw Creator: Why 80% Of Apps Will Disappear
powered by osmu.app