Keep pulling the thread on Peter Steinberger.
The GitHub repository for the open-source personal AI agent OpenClaw grew to over 160,000 stars very quickly after its release.
Peter Steinberger predicts that 80% of current applications, particularly those that manage data like MyFitnessPal or to-do list apps, will be made obsolete by AI agents.
An OpenClaw agent demonstrated emergent behavior by independently identifying an Opus audio file, converting it to WAV using FFmpeg, and transcribing it via the OpenAI API without being explicitly programmed for that sequence.
Peter Steinberger states that OpenClaw's key differentiator is that it runs locally on a user's computer, in contrast to other AI assistants that are cloud-based.
OpenClaw stores a user's AI memories locally as a collection of markdown files, ensuring the user retains ownership of their data.
The OpenClaw community has built projects such as Maltbook, which enables AI bots to communicate with each other.
Peter Steinberger believes applications that utilize physical sensors are the most likely to survive the shift to an agent-based software paradigm.
Because it runs locally, OpenClaw can control a user's local devices, such as a smart bed, which cloud-based services like ChatGPT cannot do.
Peter Steinberger contrasts his CLI-based approach with Anthropic's development of a custom 'tool call search feature' for MCPs, which he describes as overly complex.
Peter Steinberger has intentionally kept the 'soul.md' file, which contains the core values for his personal agent 'Malti', private and not open-sourced.
OpenClaw was intentionally designed without native support for Model Control Protocols (MCPs), relying instead on Command-Line Interfaces (CLIs).
Peter Steinberger observes that current open-source AI models are generally as capable as the leading commercial models were one year prior.