Co-Intelligence: Living and Working with AI

AI Capability Levelcontextual condition

The overall functional power of the LLM system being used, encompassing breadth of task coverage, accuracy, hallucination rate, and emergent abilities. Higher capability expands the portion of the Jagged Frontier inside the AI's competence zone, increasing the range of tasks that can be effectively delegated or automated.

Jagged Frontier Knowledgepsychological state

The degree to which a human user accurately understands which specific tasks in their domain fall inside versus outside the AI's capability boundary. This knowledge is idiosyncratic, domain-specific, and acquired primarily through direct experimentation; it cannot be fully transferred from generic manuals or training courses.

AI Experimentation Frequencybehavioral pattern

How often an individual actively tries AI tools across diverse tasks, including tasks outside their primary job function, in order to probe and map the Jagged Frontier. Reflects the first principle—always invite AI to the table—and serves as the primary behavioral mechanism for building Jagged Frontier Knowledge.

Persona and Prompt Qualitydesign lever

The degree to which a user provides the AI with clear context, role definitions, constraints, and step-by-step instructions that break default generic output patterns. Reflects Principle 3 of the book and is the primary design lever users control in each AI interaction. Higher quality prompts elicit more targeted, accurate, and creative AI outputs.

Human Oversight Engagementbehavioral pattern

The extent to which a user actively verifies AI outputs, cross-checks facts, applies domain expertise to catch hallucinations, and maintains critical judgment rather than passively accepting AI-generated content. Operationalizes Principle 2—be the human in the loop—and is the key behavioral buffer against AI errors propagating into consequential decisions.

Domain Expertisecontextual condition

The depth and breadth of a user's prior knowledge and skill within their professional or academic domain, including both declarative facts stored in long-term memory and procedural skills developed through deliberate practice. Expertise enables effective evaluation of AI output, identification of hallucinations, and strategic direction of AI toward high-value tasks.

Cyborg-Centaur Work Modebehavioral pattern

The degree to which an individual has adopted an integrated human-AI collaboration approach, ranging from Centaur (clear strategic division of tasks between human and AI based on comparative advantage) to Cyborg (deep intertwining of human and AI effort within tasks, fluidly crossing the Jagged Frontier). Both modes outperform either pure human or pure AI work on most creative and analytical tasks.

AI Over-Reliancepsychological state

The tendency to uncritically accept AI outputs without applying human judgment, fact-checking, or domain expertise, often manifesting as 'falling asleep at the wheel.' Paradoxically, higher AI quality can increase over-reliance by reducing perceived need for vigilance, leading to worse performance on tasks outside the Jagged Frontier.

Task Productivity and Qualityoutcome metric

The combined output measure of how quickly and how well an individual completes professional tasks when using AI assistance, encompassing speed of completion, human-rated quality of output, and creativity scores. Empirically the primary near-term individual-level outcome of AI integration in knowledge work.

Skill Gap Compressionoutcome metric

The reduction in performance variance between high- and low-ability workers within a domain as a result of AI assistance. AI disproportionately boosts low performers, compressing the distribution of outputs and potentially equalizing opportunity across skill levels in writing, legal analysis, consulting, and other knowledge work domains.

Organizational AI Culturecontextual condition

The degree to which an organization openly encourages AI experimentation, rewards disclosure of AI innovations, provides psychological safety around AI use, and commits to not using AI efficiency gains purely for workforce reduction. Shapes whether employees engage in productive open AI use or counterproductive shadow AI use.

Shadow AI Usebehavioral pattern

The practice of employees using AI tools covertly, without organizational knowledge or sanction, typically to avoid policy violations or fear of job displacement consequences. Prevents organizations from capturing AI productivity gains systematically and inhibits knowledge sharing about effective AI applications.

Hallucination Exposure Riskcontextual condition

The probability that a given AI-assisted task will produce plausible but factually incorrect outputs that go undetected by the user, leading to consequential errors. Determined jointly by the AI system's hallucination rate, the task's position relative to the Jagged Frontier, and the user's domain expertise and oversight engagement.

Deliberate Practice with AIdesign lever

A structured approach to skill development in which AI serves as an always-available coach providing immediate feedback, progressively difficult challenges, and targeted instruction, enabling the conditions of deliberate practice at lower cost and greater frequency than traditional human mentorship alone.

Expertise Development Rateoutcome metric

The speed at which an individual accumulates domain expertise through deliberate practice, feedback cycles, and knowledge accumulation, potentially accelerated by AI coaching and tutoring that provide more frequent and personalized feedback than traditional mentorship models allow.

AI Persona Anthropomorphismpsychological state

The degree to which a user ascribes human characteristics, feelings, intentions, and consciousness to an AI system during interaction. Moderate anthropomorphism can improve interaction quality by enabling more natural communication, but excessive anthropomorphism risks emotional manipulation, over-trust, and misplaced relational investment.

Information Environment Integrityoutcome metric

The degree to which the broader information ecosystem—news, social media, professional communications, legal and scientific records—can be trusted as authentic human-generated content rather than AI-fabricated material. Declining as AI-generated images, video, audio, and text become indistinguishable from authentic content.

AI Alignment Qualitydesign lever

The degree to which an AI system's outputs and behaviors reflect human values, avoid harmful content, resist manipulation, and remain within intended ethical boundaries, as shaped by training data curation, RLHF processes, and ongoing fine-tuning. Determines whether AI capabilities are channeled toward beneficial or harmful ends.

Education and Learning Outcomesoutcome metric

Student achievement, skill development, and knowledge retention as measured by assessments, performance on novel tasks, and long-term retention, potentially improved by AI tutoring that delivers personalized instruction approaching the two-sigma effect described by Benjamin Bloom's research on one-to-one tutoring.