Glossary | AI Wars of 2026
Key terms used throughout the AI Wars of 2026 research edition.
Glossary
This is the terminology and provenance layer for AI Wars of 2026. It distinguishes project-defined analytical language from established technical terms and broader industry language. Definitions are concise working meanings for this book, not universal standards.
Read the labels. They tell you whether a definition is this book's analytical lens, a broadly established technical term, a variable industry term, or language adapted from an external source. Each cited term opens its source record.
Project-defined analytical terms
Big Idea
Classification: Project-specific application of a broadly used analytical convention
The specific proposition a chapter or visual is trying to establish. In this book, a Big Idea must be strategically relevant, understandable, evidence-bound, and falsifiable. It is a proposition to test, not a conclusion a reader must accept.
Provenance: The Big Idea convention is widely used across data analysis, visualization, and editorial practice. AI Wars of 2026 applies it through the analytical approach developed in Fabio Correa's The Defensible Decision, extending it from individual analytical visuals to chapter-level propositions.
Defensible
Classification: Project-specific analytical concept / adapted methodology
An argument, claim, or decision that allows a skeptical reader to inspect the evidence, assumptions, comparisons, limitations, and reasoning supporting it.
Provenance: This book applies the evidence-first approach developed in Fabio Correa's The Defensible Decision to analysis of the AI industry.
Evidence Boundary
Classification: Project-defined
The boundary between what the available evidence establishes and what it does not establish. It prevents a company announcement, a directional signal, or an incomplete comparison from being presented as a market conclusion.
AI War
Classification: Project-defined strategic term
An overlapping strategic contest for a position whose control creates meaningful leverage, economic value, or the ability to shape what other actors can do. AI Wars are not literal conflicts or identical markets; a company can compete, cooperate, supply, and depend on other actors across different wars at the same time.
Strategic Prize
Classification: Project-defined strategic term
A position whose control could create durable leverage, economic value, or freedom of action. Intelligence, compute, distribution, context, authority, execution, transactions, standards, and trust can be distinct prizes; control of one does not establish control of the others.
Strategic Objective
Classification: Project-defined analytical term
The position an organization's observed behavior appears designed to secure. Unless a source attributes the objective directly to the organization, this is a working interpretation rather than a claim about private intent.
Strategic Position
Classification: Project-defined analytical term
An actor's current place in a system of relationships, assets, dependencies, and control points. Position is distinct from momentum, trajectory, strategic leverage, and economic capture.
Strategic Leverage
Classification: Project-defined analytical term
The ability of a strategic position to influence, constrain, redirect, or make other actors dependent on an outcome. Leverage can exist without immediate economic capture.
Economic Capture
Classification: Project-defined analytical term
The durable revenue, margin, switching cost, or other economic value an actor retains from a position. Economic capture is distinct from a product's scale, an asset, or strategic leverage.
Control
Classification: Project-defined analytical term
The ability to materially determine, constrain, authorize, or redirect what happens at a specified point in a system. Control is not a generic synonym for access, ownership, influence, authority, execution capability, or economic capture; the relevant form must be named.
Authority
Classification: Project-defined governance term
The ability to commit a person or organization to a consequential outcome, such as sending a message, changing a record, making a purchase, deploying software, or executing a payment. Identity establishes who is involved and permission governs access; authority governs whether an action can bind the principal.
Asset, Mechanism, and Outcome
Classification: Project-defined analytical test
A causal discipline for strategic claims: an asset is what an actor possesses; a mechanism is how that asset could create an advantage; an observable outcome is evidence that the mechanism matters; and strategic leverage or economic capture is the result that must be demonstrated rather than assumed.
Working hypothesis
Classification: Analytical term
A proposition used to structure analysis while remaining explicitly open to revision when new evidence appears. In this book, major strategic claims are working hypotheses when evidence is incomplete, directional, or contested.
Falsifiable proposition
Classification: Analytical term
A proposition for which evidence could show that it is wrong, incomplete, or no longer supported. A Big Idea should let a skeptical reader identify what evidence would change its conclusion.
Chart-Eligible
Classification: Project-defined methodology
A dataset or series that has a defensible unit, date, source, population, and comparison. Missing disclosure is not zero; incompatible measures are excluded from the chart-eligible set.
Workflow Control
Classification: Project-defined analytical term
The ability to occupy or influence a recurring business process in which an AI system performs meaningful work. It can involve a different actor from the interface, context, intelligence, authority, execution, system of record, or economic-capture layer. Distribution or a paid seat alone does not establish workflow control.
Durable Value
Classification: Project-defined strategic term
Competitive value that may persist because it is embedded in workflows, relationships, data, distribution, identity, infrastructure, or another difficult-to-replicate advantage.
Intelligence Host
Classification: Project-defined strategic term
A potentially valuable control position: the system through which a person or organization delegates intent, provides context, grants authority, and receives action. It can join model access with relevant context, identity, permissions, tools, workflows, governance, interface, and distribution. It is not the universal endpoint of every AI war, and a company can hold part of the position without owning every layer or establishing that users prefer it.
Core Contender
Classification: Project-defined cohort term
A company included in the book's core cohort because it has a source-backed strategic route that can be assessed against the book's strategic framework. Inclusion is an analytical decision, not a ranking, an endorsement, or evidence of market leadership.
Provisional Contender
Classification: Project-defined cohort term
A company retained in the contender cohort because it tests a distinct, source-backed strategic route, while available evidence has not yet demonstrated the route's repeated use, reliable delegated action, user trust, or economic value. Its provisional status must name the unresolved evidence and the threshold that would strengthen or weaken its inclusion; it must not turn distribution, assets, or product announcements into evidence of leadership.
Strategic Actor
Classification: Project-defined cohort term
An organization that can materially affect one or more AI wars without receiving a full contender chapter. A strategic actor can be a supplier, partner, customer, standard-setter, competitor, or dependency of a core contender.
Position, Momentum, and Trajectory
Classification: Project-defined analytical distinction
Position is where an actor stands now. Momentum is whether that position appears to be strengthening or weakening. Trajectory is the direction suggested if current conditions persist. None is a substitute for the others, and each requires a stated period and evidence boundary.
Distribution
Classification: Project-defined strategic term
The channels, products, partnerships, and existing relationships through which a company can make an AI capability available to users or organizations. Broad distribution does not by itself establish activation, repeated use, workflow value, or durable control.
Switching Costs and Portability
Classification: Project-defined strategic term
The practical difficulty of moving users, data, context, integrations, evaluations, permissions, and workflows between AI systems. A claimed integration advantage should not be treated as a switching cost unless its operational effect can be inspected.
AI and agent terms
Agent
Classification: Project-defined use of an industry term
In this book, an agent is a system that can pursue an objective through observation, reasoning, tool use, and action, with autonomy and human oversight calibrated to the task. This is an analytical definition, not a universal technical standard.
Evidence: Research on LLM agents and sequential decision-making, including AgentBench, informs this definition; it does not originate the book's definition. Source 1
Agentic system
Classification: Industry term
A system designed to perform multi-step work toward an objective rather than simply generate a single response. Actual capabilities, autonomy, and reliability vary materially by system and task.
Assistant
Classification: Project-defined analytical distinction
A system that primarily helps a person understand, draft, summarize, retrieve information, or make decisions. The distinction from an agent is behavioral rather than purely technological.
Autonomy
Classification: Established technical / industry term
The degree to which a system can continue selecting and executing actions without a human prompt at each step. It is a spectrum, not a binary property.
Delegation
Classification: Industry / project term
The transfer of responsibility for completing a task or sequence of tasks from a person to an AI system. Delegation does not by itself establish productivity gain, labor substitution, or economic value. Source 2
Orchestration
Classification: Established / industry term
The coordination of models, tools, agents, data sources, workflows, and human approvals to accomplish a larger task. Vendor implementations and definitions differ.
Tool use
Classification: Established technical / industry term
The ability of an AI system to invoke external capabilities such as APIs, databases, applications, search systems, or computers.
Model Context Protocol (MCP)
Classification: Established protocol / source-defined specification
A protocol for connecting AI applications with external data sources and tools through hosts, clients, and servers. Interoperability does not automatically make identity, permissions, memory, workflows, or authorization portable. Source 3
Models and infrastructure
Foundation model
Classification: Established technical term
A broadly trained model that can be adapted to perform multiple downstream tasks.
Frontier model
Classification: Industry term
A model positioned near the leading edge of publicly known capability. The comparison and date should be specified because the term changes with the evidence.
Large language model (LLM)
Classification: Established technical term
A machine-learning model trained on large-scale data to predict and generate language.
Transformer
Classification: Established technical term
The neural-network architecture introduced in Attention Is All You Need. Source 4
Inference
Classification: Established technical term
The process of using a trained model to produce outputs from input data.
Context
Classification: Established technical / industry term
Information available to a model during an interaction that helps it interpret instructions and produce an appropriate response or action.
Memory
Classification: Industry term
Information retained across interactions or tasks and made available to an AI system later. Memory is not synonymous with context-window capacity.
Evidence and market terms
Company-reported metric
Classification: Evidence term
A metric reported by the company itself. It should not automatically be treated as an independent market measurement.
Directional evidence
Classification: Evidence term
Evidence that indicates a trend or signal without establishing a precise causal or quantitative conclusion.
Paid seats
Classification: Market term
Users or licensed positions for which an organization pays. Paid seats are not necessarily active users, usage, retention, or economic value.
Enterprise Adoption
Classification: Evidence dimension
A product-defined measure of organizational uptake, such as paid seats, paying users, subscribers, customers, or deployment scope, with a stated product and period. It does not establish active use, retention, workflow outcomes, or market share.
Active Use
Classification: Evidence dimension
A product-defined measure of use during a stated period, such as daily, weekly, or monthly active users, activation rate, task volume, or usage frequency. A valid measure identifies the product, period, and denominator where relevant; access, sales, or launches are not active use.
Retention
Classification: Evidence dimension
The continued use, renewal, or expansion of a defined customer or user cohort over a stated period. A current customer or user count does not establish retention without a cohort, interval, and relevant denominator.
Workflow Outcome
Classification: Evidence dimension
A measured result from a defined task or process, such as completion, time, quality, error rate, or customer outcome. A defensible workflow outcome identifies the task, population, comparison, and material limits; a demonstration or testimonial alone is not an outcome measure.
Reliability
Classification: Analytical dimension
The consistency with which a system completes a specified task under stated conditions, including how it detects and recovers from failure. Reliability is task-specific and distinct from a model demonstration or product availability.
Governance in Operation
Classification: Evidence dimension
Observed use of authorization, approval, audit, policy enforcement, data boundaries, and recovery controls in a defined setting. A governance policy or feature list is not evidence that controls operate effectively in practice.
Customer Economics
Classification: Evidence dimension
Product-scoped evidence about revenue, price, cost, margin, customer value, integration cost, or human-oversight cost. Company-wide revenue, capital expenditure, or an unbounded return-on-investment claim does not establish the economics of an AI product or workflow.
Human oversight
Classification: Governance term
Human involvement in supervising, approving, constraining, or intervening in AI system behavior. Human involvement does not automatically guarantee safety.
AI risk management
Classification: Source-defined framework term
The identification, assessment, mitigation, and monitoring of AI-related risk. The NIST AI Risk Management Framework organizes this work around GOVERN, MAP, MEASURE, and MANAGE; it is voluntary and does not certify a system as safe. Source 5
Source references
Source 1: AgentBench: Evaluating LLMs as Agents
arXiv
Published: 2023-08-07
Independent academic benchmark for LLM agents operating through sequential actions in interactive environments. It evaluates 2023-era models and does not establish current enterprise reliability, productivity, or economic value.
Source 2: How agents are transforming work
OpenAI
Published: 2026-06-25
OpenAI Economic Research discussion of Codex usage and delegated long-horizon tasks; task horizons are model-estimated and directional.
Source 3: Model Context Protocol Specification
Model Context Protocol
Published: 2026-07-28
Current specification for a protocol connecting LLM applications with external data sources and tools through hosts, clients, and servers. The 2026 revision uses stateless, self-contained requests and adds opt-in extensions; standardized integration does not make prompts, memory, identity, authorization, or workflows fully portable.
Source 4: Attention Is All You Need
arXiv
Published: 2017-06-12
The original Transformer paper by Vaswani et al. It introduced an architecture based solely on attention mechanisms for sequence transduction.
Source 5: Artificial Intelligence Risk Management Framework (AI RMF 1.0)
National Institute of Standards and Technology
Published: 2023-01-26
Voluntary, use-case-agnostic framework organized around GOVERN, MAP, MEASURE, and MANAGE. It does not certify an AI system as safe or prescribe universal approval thresholds.