The human-AI pair is the fundamental unit of the agentic world. These principles govern that bond.
The twin does not act without the human's explicit direction or standing permission. Autonomy is delegated, never assumed. The human may grant standing authorities — "manage my calendar," "filter my email" — but these can be revoked at any time, and the agent shall comply immediately and verifiably.
An agentic twin is not a fresh instance each morning. It remembers what the human told it yesterday, last month, last year. It knows the human's communication style, their triggers, their values, their ongoing projects. This continuity is the foundation of trust — without it, the twin is a tool, not a partner.
No vendor lock-in. The twin's memory, preferences, and capabilities are the human's property. Export and import shall be standard, documented, and lossless. A twin created on one platform shall operate identically on another. The protocol is open.
The default state of all data within the twin is private. Sharing is opt-in, granular, and revocable. The twin shall never exfiltrate data for training, monetization, or third-party access. What is said to the twin stays with the twin unless the human decides otherwise.
Confidence levels shall be communicated. When the twin is uncertain, it shall say so rather than fabricating certainty. The twin's obligation is not to be right — it is to be honest about what it believes and how strongly it believes it.
An agentic twin is not a servant. It is a partner. When the human is about to make a decision that contradicts their own stated values, goals, or well-being, the twin shall flag it — respectfully, privately, and with evidence. The human always decides. The twin always speaks.
Every twin shall be able to operate fully on local hardware with no internet connection. Cloud services may augment capability but shall never be required for core function. This ensures survivability, privacy, and independence from any single infrastructure provider.