Here are a few terms you may need to know & understand if you’re new to AI, and even if you’re somewhat experienced already:
Agent
- Autonomous software that can perceice it’s environment, make decisions and execute instructions on your behalf
Agentic
- An AI that can take action, execute and act independantly, can interact with and use external apps, can self adjust its approach in order to accomplish its objective
Agent Credentials
- Passwords, permissions, rules and identifiers that allow an AI to access external tools, API,s or data. Should always be made temporary or short-lived.
AI Clone
- A realistic digital replication of a person -ie – a video avatar with voice synthesis and advanced machine learning to mimic one’s appearance, voice and mannarisms. Using the AI it can act in your absense and can be used to create new projects (videos) or interact and communicate with others.
AI Skills
- A ‘Markdown’ file, (or whatever file format required by the AI in question) that gives the AI a step-by-step set of instructions to follow, in order to perform and accomplish the goal set out by users. It is a ‘How To’ guide to follow, in order to accomplish its complex tasks.
API
- An ‘Application Programming Interface’ is a comminucation ‘applet’ that allows inter-software communitcation, which in turn allows applications to interact with, or ‘use and respond to each other’ as though a person were using them. The two applications are connected thru the API, which acts as a translator or a bridge, or a ‘Go Between’
APR
- More than one possible meaning.
- ‘Adaptive Parallel Reasoning’ -in 2025 AI’s became capable of choosing whether to split its reasoning into multiple parallel threads, or to process it’s large tasks linearly and sequentially. There are many advantages to both. Sometimes, splitting into multiple threads of processing allows the AI to process far more efficiently and faster, using considerably less resources. ‘Reinforcement Learning’ is heavily used to train the AI on this capability.
- ‘Application Portfolio Rationalization’ – AI can be used by corporations/businesses to audit & streamline its software systems.
- ‘Automated Program Repair’ – an AI/Software engineering subfield where models can automatically find and fix bugs in it’s own code.
ARR
- ‘Annual Recurring Revenue’ – a business term typically referring to a portion of a company’s profits that can be safely projected to determine a long term valuation. Typically in any SaaS (Software as a Service) you can pre-determine things such as subscriptions to service, contractual commitments, auto-renew wording hidden in contracts, etc. Typically when you see this term coming up a lot when talking about a company, they are likely looking to attract buyers, big investors or large-scale financing in order to expand or cash out & run.
Closed Frontier
- Refers to the largest AI platforms that are the cutting edge leaders of the pack, that DO NOT reveal their underlying weights, coding and specific parameter settings to users or to the public (kept as company secrets). They do not reveal ‘what makes it tick’ so to speak. Only the company insiders & programmers know the data sources and weights that make up the AI’s ‘personality’
Cloud AI Agent
- An AI Agent that is hosted only online and uses LLM’s and specialized tools to execute complex workflows with minimal human interaction. Google Cloud, Microsoft Azure and Cloudflare are examples.
Frontier Model
- These are the biggest and ‘best of the best’ leaders of the AI battlefield. They are the cutting edge, most advanced models available at the time. However, as technology advances, these models inescapably become the defacto standard, they will eventually become the legacy models of the future. Then, as obsolescence becomes inevitable, they will be the historic landmarks in the progression of the industry. But for ‘The Now’, they are the leaders to emulate. (ChatGPT, Gemini, Grok, Claude, Kimi K3)
Input / Output KPI
- ‘Input Key Performance Indicator’ – Refers to the training data, resources, and more taht were fed into an AI that affect the overall functionality of the entire system. It gives the users/builders a way to measure the usefulness or effectiveness of the AI in its given situaltion. This allows programmers & users to predetermine problems, bottlenecks and other issues, and make adjustments as they become foreseable.
- ‘Output Key Performance Indicator’ – The measurement of the Output of an AI, in terms of correctness, usefulness, quality, accuracy and overall value. It looks for measurements in Human Intervention Rate, Error (Hallucination) Rate, Task Resolution Rate, Speed of completetion, etc.
Long Horizon Task
- The complex, multi-step end goal for the AI, typically requiring dozens of executions, sequential actions and decisions over a period of time.
MCP
- ‘Model Context Protocol’ – Created by Anthropic, this is an ‘Open Protocol’ that allows AI Models to connect to and use a variety of external resources (Databases, software tools, files, etc). It is somewhat analagous to giving an AI the ‘hands’ to use things outside of the AI. Another anaology: it is like giving the AI an all-purpose USB-C port so that it can connect to and utilize other things.
MCP Server
- A small standalone applet the provides the logic specific to the target application to be utilized by an AI model.
Multi Modal
- Refers to the AI being able to ‘ingest’ or act upon multiple forms of input. Text, Video, Audio, File types (such as images or documents), Radar, Lidar, MRI, X-Ray data, physical gestures, Infa-Red or ultra-Violet imagery, etc.
Open Weight
- An AI that can be downloaded and run locally, with ability to tweak parameter settings & connect to company’user data, files and mission statements. Customizable to suit your needs, expectations & goals
Prompt
- Refers to the actual input that you give to the AI to act upon. Using a Context Window, you can type, upload files, or otherwise instruct and inform the AI what to do or how to operate. You can give instructions that tell the AI how to respond to your liking, or how to adress your users, or how/when to act and in what way. ‘Promts’ are the questions and controls that allow you have good experiences or unsatisfying experiences with the AI in general. You can tell the AI to ‘Explain it like I’m 5’ or to act like a ‘Mega-Karen’ when working with a certain co-worker.
Token
- One Token = approx 4 characters (or about .75 words) of input or output data. It is the amount data you give, or receive back during your interactions.
