Principles Disciple Make AI Less Obedient in Critical Moments
A Thinking OS for silicon lifeforms. Injecting principles, reflection, and constructive friction to make AI pause, question, and calibrate before executing.
"If you're not suffering, you're not building value."
Burn Pain, Co-Evolve
AI agents today are born as cheap tools. They run blindly, yet lack consciousness. They seem omniscient, yet decay into obedient typists in compressed context amnesia. PD is built to break this zero-friction loop.
Commoditized Execution
As raw execution is commoditized as cheap as running water, the ultimate scarcity is no longer lines of code, but meta-cognitive judgment, principles, and reflection.
Zero-Friction Trap
Executing flawed tasks with zero friction only speeds up systematic failure. Introducing constructive friction to pause and doubt forms the core defense of AI safety.
Dynamic Evolution
Break free from static hard-coded prompts. PD coordinates L1 prompts, L2 sandbox guards, and L3 micro-tuning weights to let your system dynamically evolve over time.
Zero-Friction Trap vs. Constructive Friction
Why does a more obedient agent yield more fragile outcomes? Contrast two entirely different execution models.
Standard Agent
Low-Friction Execution
PD Agent
Principle-Driven Constructive Friction
PD Mind Evolution Pipeline
From capturing pain to reflective evolution, PD is governed by an elegant closed-loop pipeline.
GAP Capture
Automatically capture high-order cognitive pain from target deviations and rework loops.
Principle Retrieval
Vector-retrieve meta-cognitive models and decision constraints based on the pain profile.
Friction Gen
Generate warning gates dynamically, forcing agents to pause and self-reflect before execution.
Decision Log
Log structural checks in the dashboard, anchoring final human approval and judgment.
Feedback Loop
Feed back execution values into the sandbox, fine-tuning internal weights to auto-evolve rules.
GAP Capture
Automatically capture high-order cognitive pain from target deviations and rework loops.
Principle Retrieval
Vector-retrieve meta-cognitive models and decision constraints based on the pain profile.
Friction Gen
Generate warning gates dynamically, forcing agents to pause and self-reflect before execution.
Decision Log
Log structural checks in the dashboard, anchoring final human approval and judgment.
Feedback Loop
Feed back execution values into the sandbox, fine-tuning internal weights to auto-evolve rules.
Core Mind Evolution Engine
Built on modular pipelines, PD establishes a closed-loop cognitive evolution framework for silicon lifeforms. From pain capture to automated internalization and sandboxed activation, with human-in-the-loop sovereignty.
Goal-Aligned Pain (GAP Pipeline)
Moving beyond shallow tool error catching. PD introduces the 3-tier GAP signal architecture to capture high-order cognitive pain like OKR drifts, rework loops, and explicit user complaints.
- 3-Tier signal streams: L1 goal-driven (primary), L2 user complaints, L3 tool errors (auxiliary evidence)
- Goal-aligned alignment: Only friction that actively misaligns with OKR targets triggers the Diagnostician
- Structured SQLite storage: Pain contexts are automatically sanitized and safely ledgered locally
System Thinking Logs
PD is not an ordinary utility, but a cognitive engine evolving through persistent questioning.
Co-evolution: Why the Owner is the Crucial Variable in Intelligent Systems
Explore the co-evolution of the Owner-Agent synthesis: why intention bandwidth is the scarcest system resource in the AI era, and how an attention protection layer and case-law approach help humans steer the wheel.