Every acronym and specialized term used in this document, defined in one place for quick lookup. Terms are also defined inline at their first use in the running text; this glossary is the reference copy. - **A2A (Agent2Agent):** An open protocol, originally developed by Google and now governed by the Linux Foundation, that lets independent AI agents discover each other's capabilities and communicate to delegate and coordinate tasks. - **ABAC (Attribute-Based Access Control):** An access-control model that grants permissions based on attributes of the user, resource, and context, rather than fixed roles. - **ACP (Agentic Commerce Protocol):** An emerging standard for agent-initiated commerce transactions, alongside AP2, still consolidating as of this writing. - **AI (Artificial Intelligence):** Used throughout this document as an umbrella term; specific mechanisms (agents, LLMs, RAG, and so on) are defined individually where precision matters. - **AP2 (Agent Payments Protocol):** A protocol for agent-initiated payment transactions. - **API (Application Programming Interface):** A defined interface that lets one piece of software call another's functionality directly, without a user interface. - **APIM (API Management):** Infrastructure for managing, securing, and monitoring API traffic, including AI gateway traffic. - **APM (Application Performance Monitoring):** Tooling that tracks the health, latency, and errors of running software systems. - **AUP (Acceptable Use Policy):** A published policy telling employees what AI tools they may use, on what data, and under what constraints. - **AWS (Amazon Web Services):** Amazon's cloud computing platform. - **B2B (Business-to-Business):** Commerce or transactions conducted between businesses rather than with individual consumers. - **B2C (Business-to-Consumer):** Commerce or transactions conducted between a business and individual consumers. - **BCG (Boston Consulting Group):** A management consulting firm cited for AI value-tracking research. - **BIRD-SQL (Big Bench for Large-Scale Database Grounded Text-to-SQL Evaluation):** A benchmark for evaluating a model's ability to translate natural language into correct SQL queries. - **BM25 (Best Matching 25):** A keyword-ranking algorithm used in search and hybrid retrieval systems. - **BPM (Business Process Management):** A discipline and toolset for modeling, automating, and improving structured business processes. - **CD (Continuous Delivery/Deployment):** The practice of automatically preparing or releasing every validated code change to production; paired with CI as "CI/CD." - **CFO (Chief Financial Officer):** The executive responsible for an organization's financial management. - **CI (Continuous Integration):** The practice of frequently merging and automatically testing code changes; paired with CD as "CI/CD." - **CISA (Cybersecurity and Infrastructure Security Agency):** The U.S. federal agency responsible for cybersecurity and infrastructure protection guidance. - **CISO (Chief Information Security Officer):** The executive responsible for an organization's information security program. - **CRM (Customer Relationship Management):** Software systems that manage an organization's interactions with current and prospective customers. - **CUA (Computer-Using Agent):** An AI agent that acts directly on a computer's graphical interface or browser, viewing screenshots and issuing mouse/keyboard actions, rather than calling structured APIs. - **CVE (Common Vulnerabilities and Exposures):** A public catalog of disclosed software security vulnerabilities, each with a unique tracking identifier (e.g., CVE-2025-6514). - **DLP (Data Loss Prevention):** Controls that detect and block sensitive data from leaving an organization's systems, including through AI prompts and outputs. - **DORA (DevOps Research and Assessment):** A Google-affiliated research program that produces the industry's leading evidence-based research on software delivery performance and, since 2024-2025, AI-assisted engineering. - **EDR (Endpoint Detection and Response):** Security tooling that monitors endpoint devices for threats and can trigger automated response actions. - **ERP (Enterprise Resource Planning):** Integrated software systems (finance, HR, supply chain, and similar) that run an organization's core operational processes. - **EU (European Union):** The political and economic union whose AI Act is a primary regulatory reference point in this document. - **FINOS (Fintech Open Source Foundation):** A Linux Foundation project developing open-source standards and frameworks for financial services technology, including agent governance frameworks. - **FSE (Foundations of Software Engineering):** A leading academic software engineering conference, cited here for peer-reviewed research on AI-assisted code migration. - **GCP (Google Cloud Platform):** Google's cloud computing platform. - **GenAI (Generative AI):** AI systems that generate new content (text, code, images, and so on) rather than only classifying or scoring existing content. - **GPAI (General-Purpose AI):** A regulatory category under the EU AI Act covering foundation models not built for a single narrow purpose. - **GPT (Generative Pre-trained Transformer):** OpenAI's family of large language models. - **GUI (Graphical User Interface):** A visual, point-and-click interface, as distinct from a command-line or programmatic interface. - **HHEM (Hughes Hallucination Evaluation Model):** Vectara's open model for scoring whether a generated answer is factually grounded in its source material. - **HR (Human Resources):** The organizational function responsible for workforce management. - **HTTP (Hypertext Transfer Protocol):** The foundational protocol of the web, referenced here as a comparison point for newer agent-interoperability protocols. - **IAM (Identity and Access Management):** Systems and policies that govern who (or what) can access which resources. - **ID (Identity/Identifier):** Used in this document both generically and in named products (e.g., Microsoft Entra Agent ID). - **IEC (International Electrotechnical Commission):** The international standards body that co-publishes standards such as ISO/IEC 42001 and ISO/IEC 27001 with the International Organization for Standardization. - **IEEE (Institute of Electrical and Electronics Engineers):** A professional association that publishes technical standards and research, including the ISTAS conference cited in this document. - **IP (Intellectual Property):** Legal rights in creative and inventive work; a live risk area for AI-generated content and training data. - **ISO (International Organization for Standardization):** The international standards body behind standards such as ISO/IEC 42001 (AI management systems) and ISO/IEC 27001 (information security). - **ISTAS (International Symposium on Technology and Society):** An IEEE-affiliated conference, cited here for research on security degradation in iterative AI code generation. - **IT (Information Technology):** The organizational function responsible for technology infrastructure and systems. - **JIT (Just-in-Time):** In this document, just-in-time access: granting elevated permissions only for the duration of a specific approved action. - **JSON (JavaScript Object Notation):** A lightweight, structured data format commonly used for API requests, responses, and model output schemas. - **KPI (Key Performance Indicator):** A measurable value used to track progress toward a goal; referenced generically in discussions of measurement discipline. - **L1-L4 (Maturity Levels):** The four-stage maturity scale (Experimental, Managed, Scaled, AI-Native) used throughout the Enterprise AI Capability Model to score organizational capability. - **LLM (Large Language Model):** A machine learning model trained on large volumes of text to generate and reason over natural language; the technology underlying most modern AI agents and assistants. - **MCP (Model Context Protocol):** An open standard, now governed by the Linux Foundation's Agentic AI Foundation, for connecting AI models to external tools and data sources in a consistent way. - **METR (Model Evaluation & Threat Research):** An independent research organization known for a randomized controlled trial measuring real-world AI coding productivity effects. - **MIT (Massachusetts Institute of Technology):** Cited here as the affiliation behind the NANDA research initiative on enterprise AI pilot outcomes. - **M365 (Microsoft 365):** Microsoft's cloud productivity suite (Office apps, Copilot, and related services). - **MTTR (Mean Time to Restore):** A DORA software delivery metric measuring how long it takes to restore service after a failure. - **NANDA (Networked Agents and Decentralized AI):** An MIT research initiative studying enterprise AI agent adoption and outcomes. - **NBER (National Bureau of Economic Research):** A U.S. economic research organization, cited here for field-experiment research on AI productivity effects segmented by worker skill level. - **NHI (Non-Human Identity):** A distinct, verifiable identity issued to an AI agent or automated workload, separate from any human user's credentials. - **NIST (National Institute of Standards and Technology):** The U.S. federal agency behind the AI Risk Management Framework (AI RMF) and its Generative AI Profile, widely used as reference vocabulary for AI governance. - **NLP (Natural Language Processing):** The field of computing concerned with processing and generating human language. - **NSA (National Security Agency):** The U.S. federal agency that co-published, with CISA, security guidance on Model Context Protocol risk. - **OAuth:** An open standard for delegated authorization, used (in its 2.1 revision) as the authorization model underlying several agent-to-agent and agent-identity protocols discussed in this document. - **OWASP (Open Worldwide Application Security Project):** A nonprofit foundation that publishes widely adopted, community-developed application security guidance, including the LLM/Generative AI security Top 10 referenced throughout this document. - **PDF (Portable Document Format):** A fixed-layout document file format, referenced here for source materials such as the NIST AI RMF. - **PII (Personally Identifiable Information):** Data that can identify a specific individual, a primary concern for data loss prevention and privacy controls. - **PR (Pull Request):** A request to merge a proposed code change into a shared codebase, subject to review. - **RAG (Retrieval-Augmented Generation):** Giving a model access to an organization's own documents and data at query time, rather than relying only on what it learned in training. - **RAGAS (Retrieval Augmented Generation Assessment):** An open-source framework for automatically evaluating RAG system output quality. - **RBAC (Role-Based Access Control):** An access-control model that grants permissions based on a user's assigned role. - **RMF (Risk Management Framework):** In this document, most often NIST's AI Risk Management Framework, a voluntary structure for identifying and managing AI risk across a system's lifecycle. - **ROI (Return on Investment):** A measure of the value gained from an investment relative to its cost. - **RPA (Robotic Process Automation):** Software "bots" that automate repetitive, rule-based digital tasks by mimicking user actions on existing interfaces. - **SaaS (Software as a Service):** Software delivered and licensed as a hosted, subscription-based cloud service rather than installed on-premises. - **SAP:** A major enterprise software vendor, referenced here for its Joule AI agents embedded across ERP functions. - **SAST (Static Application Security Testing):** Automated analysis of source code to find security vulnerabilities without executing the program. - **SDK (Software Development Kit):** A packaged set of tools, libraries, and documentation for building on a given platform or protocol. - **SDLC (Software Development Lifecycle):** The end-to-end process of planning, building, testing, deploying, and maintaining software. - **SEO (Search Engine Optimization):** Techniques for improving a page's visibility in search results; referenced in this document only as an excluded source-quality category. - **SOC (Security Operations Center):** The team and systems responsible for monitoring and responding to security incidents. - **SPIFFE (Secure Production Identity Framework for Everyone):** An open standard for issuing verifiable workload identities, increasingly extended to cover AI agents. - **SPIRE:** The reference runtime implementation of the SPIFFE standard. - **SQL (Structured Query Language):** The standard language for querying and managing relational databases. - **UI (User Interface):** The means by which a person interacts with a software system. - **US (United States):** Referenced primarily in the context of U.S. federal agencies and government reference frameworks (NIST, NSA, CISA). - **UX (User Experience):** The overall experience and usability of a product or system from the user's perspective. - **VM (Virtual Machine):** An isolated, software-based emulation of a computer system, commonly used to sandbox agent actions. ## Additional Terms - **Agent (AI agent):** A system that perceives its environment, reasons about what to do, takes an action through a tool, and observes the result, repeating until a task is complete. - **Agentic AI:** AI systems built around the agent loop, as distinct from single-turn chat or classification systems. - **Context engineering:** The discipline of designing the whole system around a model (what data it retrieves, what tools it can call, what history it carries), rather than focusing narrowly on prompt wording. - **Frontier model:** The current highest-capability general-purpose models from the major AI labs. - **Groundedness:** Whether a model's answer is actually supported by the retrieved source material, as distinct from output that is merely fluent and confident. - **Hallucination:** Fluent, confident, but factually incorrect model output. - **Hub-and-spoke (operating model):** An organizational pattern in which a central team owns shared platform and governance, while business-unit "spoke" teams own delivery. - **Prompt injection:** An attack in which instructions hidden in content an AI processes hijack its behavior; direct if the user does it, indirect if a third party embeds it in data the AI later reads. - **Zero-trust (for agents):** Treating every agent as potentially compromised and verifying its actions continuously, rather than trusting it once it is inside the organization's perimeter.