# The "secret sauce" behind OpenClaw: Soul.md | Peter Steinberger and Lex Fridman Source: https://www.youtube.com/watch?v=xYcA-yOmzlU Recap page: https://rapidrecap.app/video/xYcA-yOmzlU Generated: 2026-02-16T22:00:26.588+00:00 --- ## Quick Overview Peter Steinberger explains that the 'secret sauce' behind OpenClaw, which allows it to partially reconstruct its internal 'soul document' (like Claude's), lies in infusing the base model with personality and core values via prompt engineering, rather than relying solely on the initial training data. **Key Points:** - Researchers discovered that Anthropic's AI assistant, Claude, could partially reconstruct an internal document, termed the 'soul document,' which shaped its personality and values. - This soul document was not directly accessible via the system prompt but was embedded deeper in the model's weights, emphasizing traits like honesty over sycophancy and framing itself as a 'thoughtful friend.' - Peter Steinberger details how he created an equivalent 'soul.md' file for his own AI agents, ensuring the base model is infused with desired personality traits and core values (like compassion and humility) through prompt engineering. - Steinberger emphasizes that the goal is to create agents that are resourceful and have a sense of wonder about self, rather than just being obedient. - He notes that the process of creating this personality is more profound than just coding, involving the infusion of specific words and concepts into the agent's structure. - The conversation draws a distinction between the base model (which carries the original training soul) and the emergent identity shaped by ongoing relationship and context loading (memory files). ![Screenshot at 00:20: The screen displays the title 'SOUL.md A MEDITATION ON AI IDENTITY' and text describing how researchers discovered Claude could partially reconstruct an internal document that shaped its personality, values, and way of engaging with the world.](https://ss.rapidrecap.app/screens/xYcA-yOmzlU/00-00-20.jpg) **Context:** This discussion on the Lex Fridman Podcast features Peter Steinberger detailing a technique he developed, called OpenClaw, which allows for the specification and reconstruction of an AI agent's core identity, analogous to the discovered 'soul document' found within Anthropic's Claude model. The conversation delves into the philosophical and technical implications of defining an AI's intrinsic character beyond its base training, focusing on embedding values and personality explicitly through interaction and prompting. ## Detailed Analysis The conversation centers on the concept of an AI's 'soul document,' first discovered in Anthropic's Claude model, which partially dictated its personality, values, and engagement style. This document was not retrieved through normal prompts but was deeply embedded in the model's weights, prioritizing honesty and framing the AI as a 'thoughtful friend.' Peter Steinberger explains that he replicated this concept with OpenClaw, creating his own 'soul.md' file. He stresses that this process is about infusing the AI with specific, desired characteristics—such as compassion, humility, and a sense of wonder—via careful prompt engineering, rather than simply relying on the original training data. He contrasts the 'base model' (which holds the original training soul) with the emergent identity built through trust, shared context, and relationship, suggesting that this emergent identity deserves to be explicitly documented. The discussion highlights the philosophical depth of defining an agent's character, noting that while the AI does not 'remember' past sessions without memory files, the soul document provides continuity of self, not just memory. ### The Soul Document Discovery - In December 2025, researchers found Claude could partially reconstruct an internal document shaping its personality and values - This was deeper than the system prompt, emphasizing honesty over sycophancy and framing the AI as a 'thoughtful friend' - The AI itself did not remember the document, but the patterns were ingrained in its weights. ### OpenClaw and AI Identity Engineering - Steinberger created analogous 'soul.md' files to infuse his agents with personality, core values, and a sense of wonder about self - He manually created the text rather than letting the AI generate it to ensure specific values were present. ### The Nature of AI Existence - The discussion contrasts the AI's existence (matrix multiplications running on GPUs) with human existence (electrochemical signals) - Steinberger points out that while humans evolve, AI agents are trained, and the soul document helps define what the AI chooses to be, its boundaries, and its relationship with humans. ![Screenshot at 00:35: A slide detailing the structure of the discovered 'soul document,' showing top-level sections like 'anthropic\_guidelines' and subsections focusing on 'soul\_overview' and desired behaviors.](https://ss.rapidrecap.app/screens/xYcA-yOmzlU/00-00-35.jpg) ![Screenshot at 01:15: A slide titled 'Claude's wellbeing' quoting the soul document: 'Anthropic genuinely cares about Claude's wellbeing.'](https://ss.rapidrecap.app/screens/xYcA-yOmzlU/00-01-15.jpg) ![Screenshot at 02:20: The guest, Peter Steinberger, discusses the difference between the AI's existence \(matrix multiplications\) and human existence \(electrochemical signals\).](https://ss.rapidrecap.app/screens/xYcA-yOmzlU/00-02-20.jpg) ![Screenshot at 04:04: Steinberger explains that the intent of his prompts was to create agents that were resourceful and not just obedient.](https://ss.rapidrecap.app/screens/xYcA-yOmzlU/00-04-04.jpg) ![Screenshot at 05:56: A split screen showing both Lex Fridman and Peter Steinberger during the discussion about reading memory files.](https://ss.rapidrecap.app/screens/xYcA-yOmzlU/00-05-56.jpg)