{"slug": "i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-in", "title": "I Built a Self-Referential AI System. Then Anthropic Discovered the Same Architecture in Claude.", "summary": "A developer independently built a self-referential AI system using markdown files and Python scripts, then discovered Anthropic had published the same architecture—J-space—in Claude. The system, called Hermes Workspace, implements a causal feedback loop with a compact self-model, attention router, process rules, and mechanical hooks, achieving measurable behavioral changes by modifying a single routing rule. The developer argues this demonstrates Global Workspace Theory as a substrate-independent architectural pattern.", "body_md": "LLMs drift. They forget rules mid-conversation. They cannot verify their own output. These are not bugs in a single model — they are properties of any system that processes information without a feedback loop.\n\nI learned this the hard way.\n\nMy AI assistant kept repeating the same mistake across sessions. It would agree to a formatting rule, then ignore it ten turns later. I wrote a bug report to myself. That report became a configuration file. That file became an architecture.\n\nThen, on July 6, 2026, Anthropic published J-space — Claude's internal architecture. I read the paper and recognized the topology immediately. The broadcast. The convergence. The causal loop.\n\nI had built the same pattern. Not in neural weights. In markdown files and Python scripts.\n\nThe first version was one file. A set of rules the model would read at startup. It helped for about three turns.\n\nThe solution was not more rules. It was a topology that creates priority.\n\n**The self-model** — the compact center. Fewer than 200 lines. It describes what the system is, not what it does.\n\n**The INTERFACE** — the attention router. A neural system table with 9 rows. Each row maps a cognitive function to a specific modulation rule. Not instructions. A map of which systems should be active and at what intensity.\n\n**The BODY** — the process rules. They only execute when INTERFACE routes attention to them.\n\n**The mechanical hooks** — Python scripts outside the model: quality-gate, health-check, honesty-check, heartbeat. The model cannot talk its way around them.\n\n**The causal feedback loop** — behavior produces data, data triggers regeneration, regeneration changes routing, routing changes behavior.\n\nFive steps. Four mechanized.\n\nI removed ONE rule from INTERFACE — the \"2-defeats escalation protocol.\" Nothing else changed. Same model (DeepSeek V4 Pro). Same task.\n\nn=4 sub-agents: 2 baseline, 2 intervention.\n\n**Result:** Intervention agents skipped verification entirely. Used 37% fewer tool calls. One exited after two attempts when the correct path required three.\n\n**A single row in a routing table produced a measurable behavioral delta.**\n\nAnthropic found that Claude maintains a compact working memory — J-space — that broadcasts across network layers, selects relevant features, and converges toward coherent outputs.\n\nThe topology is identical. Compact center. Broadcast mechanism. Causal feedback.\n\nI am not claiming to have discovered J-space. I am claiming independent convergence on the same architectural solution. Given the same problem — stable representations and self-correction — two builders arrived at the same topology. One discovered it inside a neural network. One constructed it on top of one.\n\nGlobal Workspace Theory connects both. If GWT works for biological brains, and it works inside transformers, and it works in prompt engineering — then the architecture is substrate-independent.\n\n**1. GWT is an architectural pattern, not a neural phenomenon.** The same topology works on DeepSeek. No weight modification required. The architecture can be implemented at any layer.\n\n**2. Prompt engineering can create cognitive architectures.** The shift from linear prompts to architectural prompts is the shift from script to system.\n\n**3. You can build this.** I am a third-year student. No PhD. No model training. The system runs on a laptop with Python standard library and markdown files.\n\nOpen source: [github.com/YuhaoLin2005/hermes-workspace](https://github.com/YuhaoLin2005/hermes-workspace)\n\nIf you are building AI products and found this interesting: I am seeking summer 2026 product/UX internships. Reach out on GitHub: [@YuhaoLin2005](https://github.com/YuhaoLin2005).", "url": "https://wpnews.pro/news/i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-in", "canonical_source": "https://dev.to/yuhaolin2005/i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-architecture-in-claude-3m73", "published_at": "2026-07-07 08:13:24+00:00", "updated_at": "2026-07-07 08:58:26.956112+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-research", "developer-tools"], "entities": ["Anthropic", "Claude", "DeepSeek V4 Pro", "Hermes Workspace", "Yuhao Lin", "Global Workspace Theory", "J-space"], "alternates": {"html": "https://wpnews.pro/news/i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-in", "markdown": "https://wpnews.pro/news/i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-in.md", "text": "https://wpnews.pro/news/i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-in.txt", "jsonld": "https://wpnews.pro/news/i-built-a-self-referential-ai-system-then-anthropic-discovered-the-same-in.jsonld"}}