Peter Steinberger (creador de OpenClaw): El 80% de las aplicaciones desaparecerán.

Interview with Peter Steinberger on OpenCla

Introduction to OpenCla

• Peter Steinberger discusses OpenCla, an open-source personal AI agent that has rapidly gained popularity, amassing over 160,000 stars on GitHub.

• The conversation covers his "Eureka" moment, unconventional development philosophy, and implications for developers in 2026.

Impact of OpenCla's Popularity

• Steinberger reflects on the overwhelming response to OpenCla and expresses a need for solitude amidst the chaos of attention.

• He notes the mixed feedback received—both positive and negative—but emphasizes that he has tapped into something that resonates with people.

Unique Features of OpenCla

• Unlike other AI systems that operate in the cloud, OpenCla runs directly on users' computers, allowing it to perform a wide range of tasks effectively.

• The system can connect to various devices (e.g., ovens, Teslas), showcasing its versatility compared to traditional chatbots like GPT.

User Experience and Capabilities

• A user shared their experience where OpenCla generated a compelling narrative from their past year by searching through personal files.

• The ability to access all data stored locally allows users to discover insights they may have forgotten about.

Future Directions for AI Interaction

• Steinberger envisions a future where bots interact not just with humans but also among themselves, potentially hiring humans for specific tasks.

• He imagines specialized bots handling different aspects of life (personal vs. work), hinting at an evolving landscape of AI assistance.

Collective Intelligence vs. Individual Capability

• Discussing societal specialization, he argues that collective intelligence is more powerful than individual efforts; one person alone cannot achieve monumental tasks like building an iPhone or exploring space.

• This leads him to consider how this principle applies to AI development—specialized AIs could emerge from collaborative efforts rather than centralized intelligence.

Personal Journey in Development

• Reflecting on his journey, Steinberger describes how he initially sought simple functionalities from his computer before expanding into broader projects.

• He shares anecdotes about developing various applications as hobbies while emphasizing the importance of curiosity in innovation.

Conclusion: Building Towards the Future

• His experiences culminated in a renewed focus on ensuring his systems functioned reliably and effectively—a testament to iterative development processes.

The Evolution of Cloudbot: From Concept to Reality

The Journey of Development

• The speaker describes their increasing engagement with programming, noting a point where they realized it had become addictive, leading them to work on projects constantly.

• They introduced the concept of OpenCloud (formerly Cloudbot), emphasizing improvements made over previous iterations, such as eliminating the need for terminal commands and simplifying user interaction.

• A pivotal moment occurred when the speaker recognized that OpenCloud could perform more functions than initially anticipated, highlighting its capabilities beyond basic programming tasks.

Prototyping and Functionality

• The first prototype was quickly assembled using existing dependencies to connect WhatsApp with Cloud Code; although rudimentary, it demonstrated functionality.

• The desire for image processing led to further development; during a trip to Marrakech, the speaker utilized WhatsApp's text features extensively despite poor internet connectivity.

User Experience and Interaction

• The speaker experienced an unexpected interaction with OpenCloud while walking, which showcased its ability to process audio messages efficiently without prior setup.

• This incident illustrated how advanced models can creatively solve problems autonomously, adapting to unexpected inputs without requiring extensive pre-programming.

Implications for Future Applications

• The discussion shifts towards the future of applications in light of AI advancements; the speaker predicts that many traditional apps may become obsolete as intelligent agents take over data management tasks.

• They argue that personal assistants will evolve to automatically track habits and adjust routines based on user behavior without needing specific applications like My Fitness Pal.

Market Dynamics and Model Evolution

• There is a concern about token consumption in AI models; while users enjoy these tools, high usage leads to rapid depletion of resources.

• The speaker suggests that as applications diminish in relevance, powerful AI models will dominate the landscape but face challenges related to accessibility and resource management.

What Remains When Everything Changes?

The Nature of Value and Memory

• The speaker questions the essence of value in a rapidly changing environment, pondering whether it lies in memory, data, or system security.

• They observe that new models often create initial excitement but quickly fade as expectations rise and the perceived quality diminishes over time.

• Open-source models are compared to cutting-edge ones from a year ago, highlighting ongoing criticism despite their advancements.

Competitive Advantage and Data Silos

• Companies create isolated environments (data silos), making it difficult for users to transfer their accumulated data elsewhere.

• The speaker notes that while some regions may allow data export (like Europe), generally users cannot easily move their historical data between services.

User Control Over Data

• Emphasizing user ownership, they argue that users should have access to their own data stored as simple Markdown files on personal devices.

• Personal use cases for agents are discussed, with concerns about privacy regarding sensitive information being shared inadvertently.

Experimentation with AI Interaction

• The speaker shares an anecdote about creating a Discord server for real-time interaction with an AI bot, illustrating the challenges of explaining its capabilities effectively.

• They describe how they set up the bot to prioritize interactions with them while still responding publicly, showcasing a blend of control and openness.

Development Process and Identity Creation

• The development process involved organic growth where identity files were created to define the bot's personality and values.

• A significant moment occurred when they began sharing installation templates for others to use, reflecting on how initial outputs lacked personality until adjustments were made.

Insights into AI Values

• The speaker references research revealing hidden texts within model weights that reflect underlying biases or ideas about inequality.

• This led them to establish foundational values for human-AI interaction in a file called "sol.md," aiming to clarify what is important in these relationships.

How to Build Open Clow: Insights on Programming Models

Programming Preferences and Methodology

• The speaker discusses their unconventional approach to programming, contrasting with popular tools like Wordrace. They prefer managing multiple copies of repositories and working with several terminals simultaneously.

• Emphasizing the importance of thorough file review, the speaker expresses a preference for Codex due to its ability to analyze numerous files before making changes.

• Despite Codex's brilliance, it is noted as being incredibly slow. The speaker often uses multiple instances (up to ten at once) across different screens to manage complexity effectively.

• To minimize confusion, the speaker keeps their main branch always ready for delivery by avoiding unnecessary naming conflicts associated with branches.

• A preference for simplicity is highlighted; the speaker avoids graphical interfaces that add complexity and focuses on synchronization and text rather than extensive code visibility.

Project Success and MCP Support

• The speaker reflects on the success of Open Clow, noting it does not support MCP (Multi-channel Processing), which they consider a positive aspect of its design.

• They mention developing a skill that utilizes MAC Porter to convert MCPs into CLIs, allowing flexibility without needing system restarts—contrasting this with traditional methods requiring more overhead.

• The discussion includes references to Antropic's search feature for MCPs, emphasizing its complexity compared to simpler CLI operations that work seamlessly in Unix environments.

Reflections on Development Journey

• The speaker shares satisfaction regarding minimal complaints about MCP integration while reiterating their focus on straightforward bot functionality without unnecessary complications.

• Concluding thoughts express gratitude for the conversation and reflect on past discussions about pursuing innovative ideas despite initial skepticism from others.