Glossary

Your lens

My own understanding of these terms, not official definitions, so that we read the Insights with the same vocabulary.

Agent: Software that uses AI to pursue a goal you set. It works towards it step after step, picking its own route and acting inside limits set in advance.
Agentic coding: An AI agent does the coding itself: given a task, it changes the code, tests it and corrects what fails. Your team's work moves to either end — stating the task clearly and judging what comes back.
Agentic Engineering: Building with agents at every stage of software development, from planning to operation, under the engineering standards a business-critical system already requires. The speed comes from the agents; the accountability stays where it was.
AX (Agent Experience): How well your own systems are built for AI agents to work in — what User Experience is for your customers, this is for the agents. A good agent experience does not happen by itself: the systems have to be tailored to agents.
Human in the loop (HITL): Automation does the work; a person still approves what cannot be undone. It is how accountability stays with a named human when the work itself is machine-made.
Non-determinism: The same request does not reliably produce the same result — two runs can differ and both be right. It is a property of the technology, not a fault in it, and a better model does not remove it.
PRD (Product Requirements Document): The statement of what the software must do for the business, written so the people who own the business rules can read and correct it before anything is built. Scope is agreed here or argued about later.
SDLC (Software Development Lifecycle): The phases software passes through, from a business need to running it in operation. An AI programme has to be scoped against all of them, because equipping one phase changes one phase.
Skill: Instructions an AI agent reads and follows to do a particular job the way your company does it. Writing them down is how a method stops leaving when the person who worked it out does.
Vibe Coding: Letting an AI model build software and judging it only by what it does. Whether the code behind it is sound is not part of the question.