The Principles
Higher-order laws and relationships.
Principles are different from terms. Terms name individual phenomena; principles describe the relationships between them. A principle is a claim about how the world works — a pattern reliable enough to predict outcomes and guide decisions. These 23 principles distill the patterns that emerge from observing agentic AI systems in practice. They're not theoretical; they're the recurring truths that show up when you build, operate, and measure real systems.
The Context Compensation Principle
As context size and entropy increase, required reasoning grows nonlinearly.
The Bleed Principle
Unmanaged state costs compound. Every session without persistent memory costs more than the last.
The Divergence Principle
Agent autonomy and human understanding are inversely correlated without active inspection.
The Amplification Principle
AI's value scales with the quality of the system around it. Better context, memory, and structure unlock nonlinear returns.
The Opacity Tax
The less visible a system's reasoning and state, the more cognitive load the human bears. Opacity doesn't eliminate complexity — it shifts it to the user.
The Compounding Principle
Knowledge that persists and compounds across sessions produces nonlinear returns. Knowledge that resets or is siloed produces linear costs.
The Transparency Threshold
Below a certain level of system transparency, human trust becomes indistinguishable from blind faith.
The Fragmentation Cost
Every additional tool, agent, or silo that doesn't share state multiplies — not adds — coordination overhead.
The Delegation Paradox
The more capable an agent becomes, the more the human delegates — and the less the human can verify.
The New Digital Divide
AI amplification creates a new generation of haves and have-nots. Access to capable AI systems — and the skill to use them — compounds advantage.
The Acceleration Problem
Rapid increases in model capability create rate-of-change that exceeds most organizations' ability to adapt.
The Intelligence Cascade
Model intelligence flows from frontier to medium to small. Today's frontier capability becomes tomorrow's mid-tier default.
The Specialization Principle
Small models don't compete with frontier models — they specialize. System architecture determines which model is appropriate, not raw capability.
The Substrate Principle
The model is the engine, not the car. The value of AI is determined by the system around it.
The Accountability Principle
Any system that acts on your behalf must be transparent and auditable. If you can't trace what happened, why, and who authored it, you don't have an agent — you have a liability.
The Separation Principle
Planning, state, memory, and evaluation should live outside the model. Models reason; systems remember, plan, and verify.
The Validation Gate
No agent output should reach production without structured evaluation against explicit criteria.
The Local-Cloud Spectrum
Not every task needs a frontier model, and not every model needs to be in the cloud.
The Context is the Product
The model is commoditizing. Context — persistent, inspectable, searchable, sharable — is the durable competitive advantage.
The Portability Imperative
Knowledge locked in one tool, one vendor, or one person is knowledge at risk.
The Recovery Tax
Failures in agentic systems have significant costs. Recovery from failures often exceeds the original task in time, tokens, and cognitive load.
The Trust Ratchet
Trust in AI systems is hard to build and easy to destroy. Each success adds incrementally; each failure subtracts dramatically.
The Efficiency Metric
The meaningful measure of AI cost is not tokens consumed but tokens required per completed task. The winning metric is tokens per task completion, not tokens per session.