# Terminology [**Agentic AI**](https://www.ibm.com/think/topics/agentic-ai)\ Agentic AI refers to autonomous systems capable of using reasoning and tools to complete complex, multi-step goals with minimal human intervention.
Video: "The Power of AI Agents and Agentic AI Explained" by IBM {% embed url="" %}
[**Agentic Workflows**](https://www.ibm.com/think/topics/agentic-workflows)\ Iterative design patterns where an AI agent decomposes a task into sub-goals, reflects on its own output, and uses tools to reach a final result.
Video: "How to Use Agentic AI: LLMs, AI Agents & Prompt Engineering in Action" by IBM {% embed url="" %}
[**AI Agents**](https://www.ibm.com/think/ai-agents)\ Autonomous or semi-autonomous software entities that use model reasoning to complete specific tasks or achieve goals with minimal human intervention.
Video: "What are AI Agents?" by IBM {% embed url="" %}
[**AI Agent Development**](https://www.ibm.com/think/topics/ai-agent-development)\ AI agent development is the process of designing and building autonomous software systems that can perceive their environment, reason through tasks, and use tools to achieve specific goals without constant human guidance. [**AI Guardrails**](https://www.ibm.com/think/topics/ai-guardrails)\ Safety mechanisms and rules embedded into AI systems to prevent the generation of harmful, biased, or off-topic content. [**AI Literacy**](https://www.ibm.com/think/insights/ai-literacy)\ The set of skills and knowledge required to understand, critically evaluate, and effectively use artificial intelligence technologies in various contexts. [**AI Orchestration**](https://www.ibm.com/think/topics/ai-orchestration)\ The coordination and management of multiple AI models, tools, and data sources to execute complex, multi-step workflows. [**AI Transparency**](https://www.ibm.com/think/topics/ai-transparency)\ The practice of making the operations, data sources, and decision-making logic of an AI system visible and understandable to users. [**Algorithmic Bias**](https://www.ibm.com/think/topics/algorithmic-bias)\ Systemic and repeatable errors in an AI system that create unfair outcomes, often stemming from skewed training data or flawed model design. [**Artificial General Intelligence (AGI)**](https://www.ibm.com/think/topics/artificial-general-intelligence)\ A theoretical type of AI that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks at a level equal to or beyond human cognitive capabilities.
Video: "8 Use Cases for Artificial General Intelligence (AGI)" by IBM {% embed url="" %}
[**Artificial Superintelligence (ASI)**](https://www.ibm.com/think/topics/artificial-superintelligence)\ A hypothetical form of synthetic intelligence that surpasses the collective cognitive performance of humans across every field, including scientific creativity, general wisdom, and social skills.
Video: "What is Artificial Superintelligence (ASI)?" by IBM {% embed url="" %}
[**Chain of Thought (CoT)**](https://www.ibm.com/think/topics/chain-of-thoughts)\ A prompting technique that encourages a model to generate intermediate reasoning steps to improve its performance on complex logic problems. [**Constitutional AI**](https://constitutional.ai/) \ A set of techniques developed by [Anthropic](https://www.anthropic.com/) researchers to align AI systems with human values to make them helpful, harmless, and honest. The key ideas behind Constitutional AI are aligning an AI's behaviour with a 'constitution' defined by human principles, using techniques like self-supervision and adversarial training, developing constrained optimisation techniques, and designing training data and model architecture to encode beneficial behaviours. [**Context Engineering**](https://community.ibm.com/community/user/blogs/philip-dsouza/2025/09/09/why-context-engineering-is-the-new-buzzword-in-ai)\ The strategic process of selecting and structuring the specific data and instructions provided to a model to ensure relevant and accurate responses. [**Context Window**](https://www.ibm.com/think/topics/context-window)\ The maximum amount of text or data a model can process and "remember" at any one time during a single interaction. [**Continuous Intelligence (CI)**](https://www.techtarget.com/searchbusinessanalytics/definition/continuous-intelligence)\ A data analysis process that delivers real-time analytics and insights as a part of ongoing business operations. [**Country of Geniuses**](https://www.darioamodei.com/essay/the-adolescence-of-technology) "Country of geniuses" often refers to a 2026 prediction by Anthropic CEO Dario Amodei regarding AI, where he envisions a "country of geniuses in a datacenter", a collection of millions of AI agents with the capabilities of top human experts. This concept suggests a future with superhuman AI agents outperforming humans in most cognitive tasks by 2027. [**Data Sovereignty**](https://www.ibm.com/think/topics/data-sovereignty)\ The principle that digital data is subject to the laws and governance structures of the nation or organisation where it is collected or stored. [**Deep Blue**](https://kotrotsos.medium.com/developers-now-have-a-name-for-what-they-are-feeling-f9fbcb2addd6)\ A term coined by Simon Willison and Adam Leventhal to describe the psychological ennui and existential dread that software developers are experiencing as AI increasingly automates their craft. [**Deep Learning**](https://www.ibm.com/think/topics/deep-learning)\ A subset of machine learning based on artificial neural networks with multiple layers that enables a system to automatically find patterns and features in complex data such as images, sound, and text.
Video: "Machine Learning vs Deep Learning" by IBM {% embed url="" %}
[**Explainable AI (XAI)**](https://www.ibm.com/think/topics/explainable-ai)\ The methods and techniques used to ensure that the internal mechanics and outputs of AI systems can be understood by human experts. [**Generative AI**](https://www.ibm.com/think/topics/generative-ai)\ A category of artificial intelligence capable of creating new content, such as text, images, or audio, based on patterns learned from existing data. [**Generative Pretrained Transformer (GPT)**](https://www.ibm.com/think/topics/gpt)\ A type of large language model architecture that uses deep learning and a transformer structure to process and generate human-like text based on patterns found in vast datasets.
Video: "Transforming Language with Generative Pre-trained Transformers (GPT)" by IBM {% embed url="" %}
[**Grounding**](https://www.gigaspaces.com/blog/grounding-ai)\ The process of linking AI outputs to verifiable facts or specific source documents to reduce errors and ensure real-world accuracy. [**Hallucination**](https://www.ibm.com/think/topics/ai-hallucinations)\ A phenomenon where an AI model generates information that sounds confident and fluent but is factually incorrect or nonsensical. [**Human-in-the-Loop (HITL)**](https://www.ibm.com/think/topics/human-in-the-loop)\ Human-in-the-loop is a process where a human provides direct feedback or intervention to an artificial intelligence system to improve its accuracy and ensure its decisions are reliable. [**Large Language Models (LLMs)**](https://www.ibm.com/think/topics/large-language-models)\ Advanced AI models trained on vast amounts of text data to understand, generate, and manipulate human language.
Video: "How Large Language Models Work" by IBM {% embed url="" %}
[**Large Reasoning Model (LRM)**](https://medium.com/@fahey_james/large-reasoning-models-lrms-an-overview-19837b72540f)\ A Large Reasoning Model (LRM) is a type of artificial intelligence designed to handle complex, multi-step tasks by using advanced internal processes to evaluate different paths and verify its own logic before providing an answer.
Video: "What Are Large Reasoning Models (LRMs)? Smarter AI Beyond LLMs" by IBM {% embed url="" %}
[**Machine Learning**](https://www.ibm.com/think/machine-learning)\ Machine Learning is a subset of artificial intelligence that uses algorithms to identify patterns in data and improve its own performance on specific tasks over time without being explicitly programmed for every outcome.
Video: "Machine Learning Explained: A Guide to ML, AI, & Deep Learning" by IBM {% embed url="" %}
[**Meta Prompting**](https://www.ibm.com/think/topics/meta-prompting)\ A technique where an initial set of instructions is used to guide the model in generating, refining, or managing its own prompts to improve the quality and structure of its final output. [**Mixture-of-Experts (MoE)**](https://www.ibm.com/think/topics/mixture-of-experts)\ A sparse neural network architecture that scales model capacity by routing specific tasks to a subset of specialised "expert" sub-networks rather than activating the entire model for every request. [**Model Context Protocol (MCP)**](https://www.ibm.com/think/topics/model-context-protocol)\ An open standard that enables seamless integration and data exchange between AI applications and various external data sources or tools.
Video: "What is MCP? Integrate AI Agents with Databases & APIs" by IBM {% embed url="" %}
Video: "MCP vs API: Simplifying AI Agent Integration with External Data" by IBM {% embed url="" %}
[**Model Tuning**](https://www.ibm.com/think/topics/model-tuning)\ The broad practice of adjusting a model's parameters or training data to improve its performance for specific tasks or domains. [**Multimodal AI**](https://www.ibm.com/think/topics/multimodal-ai)\ The capability of an AI system to process and generate information across different formats simultaneously, such as text, images, and audio. [**Narrow AI**](https://www.ibm.com/think/topics/artificial-intelligence-types)\ A specialised form of machine intelligence designed to perform a single, specific task or a limited range of functions with high proficiency, rather than possessing general problem-solving abilities. [**Neural Network**](https://www.ibm.com/think/topics/neural-networks)\ A neural network is a computational model inspired by the human brain that uses interconnected layers of nodes to process data and recognise patterns for tasks like classification or prediction. In other words, it is an artificial intelligence architecture that organises simple processing units into layers to learn the specific weights and biases needed to transform input data into accurate predictions.
Video: "Neural Networks Explained in 5 minutes" by IBM {% embed url="" %}
[**Open-Weight AI Models**](https://medium.com/lets-code-future/open-weight-ai-models-what-they-are-and-why-openais-next-move-matters-f86fe481973a)\ In machine learning, weights are the numerical parameters within a neural network that determine how input data is transformed into an output. During training, these values are adjusted to "learn" patterns and behaviours. While proprietary models keep these parameters strictly confidential, an open-weight model makes these pre-trained values publicly available for download. This enables users to perform local inference, conduct security audits, or fine-tune the model on private data without needing to send information to a third-party provider. [**Prompt Engineering**](https://www.ibm.com/think/topics/prompt-engineering)\ The craft of refining and optimising the text input given to a model to elicit the most accurate or creative output possible. Guide: "[The 2026 Guide to Prompt Engineering](https://www.ibm.com/think/prompt-engineering)" by IBM [**Prompt Tuning**](https://www.ibm.com/think/topics/prompt-tuning)\ A process that involves adding a small, trainable sequence of continuous vectors to the input of a fixed, pre-trained model to guide its behaviour for a specific task without altering the underlying parameters.
Video: "What is Prompt Tuning?" by IBM {% embed url="" %}
[**Red Teaming**](https://www.ibm.com/think/topics/red-teaming)\ The practice of rigorously testing an AI system by simulating adversarial attacks to identify vulnerabilities, biases, or safety flaws. [**Reinforcement Learning from Human Feedback (RLHF)**](https://www.ibm.com/think/topics/rlhf)\ A training method that uses human rankings of model responses to align the AI's behaviour with human values and preferences.
Video: "Reinforcement Learning from Human Feedback (RLHF) Explained" by IBM {% embed url="" %}
[**Retrieval-Augmented Generation (RAG)**](https://www.ibm.com/think/topics/retrieval-augmented-generation)\ A technique that gives an LLM access to external, trusted data sources to provide more accurate and up-to-date answers.
Video: "What is Retrieval-Augmented Generation (RAG)?" by IBM {% embed url="" %}
[**Stochastic Parrot**](https://en.wikipedia.org/wiki/Stochastic_parrot)\ Coined by Emily M. Bender and colleagues in 2021, this negative machine learning metaphor describes large language models as systems that merely mimic text through statistical patterns without possessing genuine understanding. [**Vibe Coding**](https://www.ibm.com/think/topics/vibe-coding)\ A conversational or high-level approach to programming where a user describes desired software features to an AI which then generates the underlying code.
Video: "What Is Vibe Coding? Building Software with Agentic AI" by IBM {% embed url="" %}