Software & AI engineering glossary
A practical, jargon-light glossary written by the engineers at Kiebot. Bookmark it for the next scoping call.
Software engineering
Foundational terms used in every Kiebot engagement.
- MVP (Minimum Viable Product)
- The smallest possible version of a product that delivers real value to a user and lets you learn whether the bigger idea is worth building.
- Reactive Programming
- A paradigm where systems are modelled as flows of asynchronous events, allowing components to stay responsive under load and failure. Foundational to high-throughput systems.
- Microservices
- An architecture style where an application is broken into small, independently deployable services that talk over the network. Pays off above a certain team size; expensive below it.
- API Contract
- A formal specification (OpenAPI, GraphQL schema, Protobuf) that describes what a service accepts and returns. Treat it as the source of truth for both teams and tests.
- Technical Debt
- The implicit cost of shortcuts taken today that you will pay back later in time, money, or stability. Like financial debt: cheap if managed, ruinous if compounded.
AI applications
Terms you will hear in every AI engineering conversation in 2026.
- LLM (Large Language Model)
- A neural network trained on enormous amounts of text. It generates text by predicting one token at a time, conditioned on what came before.
- RAG (Retrieval-Augmented Generation)
- A pattern where the model answers questions using documents fetched from a vector store at query time, instead of relying only on its training data.
- Vector Database
- A database optimised for storing and searching high-dimensional embeddings. Used to find semantically similar items (e.g., chunks of text that match a question).
- Embedding
- A numeric representation of a piece of text, image, or audio in a high-dimensional space. Close points mean "semantically similar."
- Prompt Engineering
- The practice of crafting and refining the instructions sent to an LLM to get reliable, structured, useful outputs. More like UX design than coding.
- Eval (Evaluation Set)
- A curated dataset of inputs and expected outputs used to measure whether a model or prompt change actually improves quality. The closest analogue to unit tests in AI.
- Agent
- A program that uses an LLM as its reasoning engine and can take actions in the real world (call APIs, write files, send messages). Reliability comes from tight tool definitions and guardrails.
- Hallucination
- When an LLM produces text that is fluent and plausible but factually wrong. The defining failure mode of generative systems.
- Temperature
- A setting between 0 and 1 that controls randomness in model output. Low for extraction or code, higher for creative writing.
- Token
- A unit of text used by LLMs. Roughly four characters of English. LLM pricing, latency, and context windows are all measured in tokens.
DevOps & cloud
The infrastructure vocabulary every engineering leader should know.
- CI/CD
- Continuous Integration (every commit is tested) and Continuous Delivery/Deployment (every passing commit can be shipped). The backbone of any high-velocity team.
- IaC (Infrastructure as Code)
- Defining your infrastructure in version-controlled code (Terraform, Pulumi, CloudFormation) instead of clicking through a console. Makes environments reproducible.
- Observability
- The ability to ask new questions about a running system without redeploying. Built on the three pillars of metrics, logs, and traces.
- SLO (Service Level Objective)
- A measurable target for the reliability of a service (e.g., 99.9% successful checkouts). What you actually commit to internally and to customers.
- Blue/Green Deployment
- A release pattern where the new version goes live next to the old one. Traffic switches over only after health checks pass; rollback is instant.
- Canary Release
- Sending a small percentage of traffic to a new version first, then gradually expanding. Limits blast radius if the release is bad.
QA & quality
How modern teams keep software working through rapid change.
- Shift Left
- Moving quality checks (testing, security review, accessibility) earlier in the development lifecycle, before they become expensive to fix.
- Regression Testing
- Re-running tests after a change to confirm that nothing previously working has broken. Best automated, never skipped.
- Test Pyramid
- A guideline that you should have many fast unit tests, fewer integration tests, and a small number of slow end-to-end tests. Optimises for feedback speed.
- Exploratory Testing
- Unscripted, intentional testing where a skilled tester learns the product, designs tests in real time, and uses judgement to find issues automation will miss.
- Flaky Test
- A test that sometimes passes and sometimes fails without code changes. Treat them as urgent bugs — they destroy trust in your CI signal.
Product & delivery
How engineering work is scoped, prioritised, and shipped.
- Scrum
- An Agile framework with short, fixed-length sprints, a product backlog, daily standups, and end-of-sprint reviews. Optimised for predictability.
- Kanban
- A continuous-flow alternative to Scrum where work moves through a board with WIP (work-in-progress) limits. Optimised for throughput and flexibility.
- Definition of Done
- The shared, written agreement of what "complete" means for a work item — including code review, tests, docs, and deployment criteria.
- Discovery
- The phase where the team validates a problem is worth solving before investing in delivery. Often the highest-leverage time spent in a product cycle.
- Retrospective
- A regular meeting where a team reflects on what worked, what did not, and what they will change next. The compounding interest of process improvement.
Want a deeper dive?
Read the long-form versions of these ideas in our engineering insights — written by practitioners who ship production code.
Browse insights