Reference

The AIOS Glossary

Every framework and coined term in the AIOS system, defined in one place — with a link to where each one lives.

AIOS: The Artificially Intelligent Operating System — the operating model and methodology for becoming AI-native. AI stops being an app you open and becomes the operating system your work runs on. The book is the practitioner manual for the model.
AI-Native: Rebuilding how work gets done around AI rather than bolting it on. An agent can find context, produce finished work to your standard, and what it learns persists.
The AI-Native Threshold: The dividing line between using AI and being organized around it — the difference between a faster horse and a car.
The Maturity Ladder: Four rungs applied to organizations and individuals: Tourist → Experimenter → Operator → AI Native. Gains don't just get bigger, they change kind.
The BASIC Stack: The five-layer framework for an AI-native baseline: Brain, Assistants, Skills, Interface, Controls — assembled in order.
The Two-Model Doctrine: Run two models on purpose, not five by accident: the assistant in your office suite for in-flow work, plus one frontier stack for serious agentic work.
Skill Chains: A 'macro' skill that fires several skills in sequence (draft → refine → quality-check), each improving the output before the next takes over.
Evals / Rubric-as-Eval: You cannot manage what you cannot see. The self-improving loop: draft → grade against a rubric → revise the weakest dimensions → re-grade until it clears your bar.
The AI-Native Flywheel: People manage agents, agents act on context, producing Speed → Signal → Moat that feeds back so the company gets smarter every loop.
Agents Are the New SaaS: Instead of renting software for every job, build agentic workflows that do the job. The target is the job, not the tool.
The Coordination Tax: The number of people who must agree to a change scales with the connections between them — which is why small teams compound faster.
Manufacture Time, Then Reinvest It: The AI-assisted org pockets the saved hours; the AI-native org reinvests the manufactured time into previously-unreachable work.
The 80-to-100 Problem: A generic model gets you an 80% draft. The last 20% — shippable, repeatable, at your standard — is where context lives.
Shadow AI: Employees using unsanctioned AI tools. A demand signal, not a crime wave — it maps the workflows that need sanctioned tooling.
The Route Audit (S.M.A.R.T.): Sort, Match, Automate, Refine, Take Control — a repeatable method for picking the workflows where AI actually pays.
The First-Route Filter: A good first workflow is high frequency, high friction, measurable in 90 days, and non-customer-facing.
Agents as Workforce 2.0: An agent is an entry-level employee. Onboard it: job description → ID badge → probation → performance management.
The Substitution Trap: Most automatable-looking jobs are two jobs in a trench coat: producing output and catching failures. AI takes the first and drops the second.
Convert First, Hire Deliberately, Never Cut First: The workforce doctrine. The converted insider — institutional knowledge plus AI fluency — is the most valuable profile in the market.
The Three-Lens Search: Every AI opportunity traces to one constraint removed: economics, speed, or scale.
The Opportunity Portfolio: Fund many cheap probes with tight windows and pre-written kill criteria. Evidence promotes; opinions don't.
Confidently Unproductive: Workers who are measurably slower with AI while believing they're faster. Measurement is the only exit.
The Four Signals: The governance framework from The Human Signal: Warning (Transparency), Voice (Participation), Veto (Accountability), Learning (Adaptation). A system that hears all four can be governed; one that hears none governs you. Warn, voice, veto, learn.
The Operating Model (The Human Signal): A seven-layer AI governance system — Threat, Build, Operate, Govern, Secure, Fund, Sustain — that only holds in concert. The governance layer that complements the AIOS operating model.
Human-in-the-Loop: Keeping a named person between the machine and any decision that can hurt someone — not to slow it down, but because they see the exception the model can't. Operationalized as the Veto Signal.