Tech Jargon Glossary: The Satirical Developer Dictionary

An A-to-Z dictionary of every overused tech buzzword you've ever winced at in a standup. Each entry is defined as it's actually used in the wild - not as the original spec intended. Pair this with our buzzword bingo card for maximum standup productivity.

Agile: The art of having two-week sprints that produce nothing while pretending the four-month delay is a feature.
AI-Powered: A regex, except the regex went to a conference.
Blockchain: A database your CTO mentioned in 2017 and has not stopped mentioning.
Cloud-Native: Runs on someone else's computer, but proudly.
Composable Architecture: Composed of vaguely working parts that each have a different opinion about JSON.
DevSecOps: DevOps, but the security team is now allowed to attend the retros they were already crashing.
Edge Computing: Compute that runs at the edge, meaning a Raspberry Pi duct-taped to a pinecone we cannot reach when it breaks.
Federated: Pretending the 50 things are one thing.
Generative AI: Confident lies, but in paragraph form.
Hallucination: What we now call "the wrong answer" so we don't have to fix it.
Idempotent: Safe to call twice. The agent calls it forty-seven times. nut-billing now owes the customer money.
Just: The word that turns a four-week task into a five-minute Slack message.
Kubernetes: A YAML factory. Our cluster is named prod-ish and reports emotional readiness.
Leverage: "Use," but in a meeting.
Microservices: A monolith that costs more to host. Ours has 147 services. Four of them do anything. Twelve call each other in a perfect circle.
Native: Built once. Re-platformed twice. The third rewrite is "agentic."
Observability: Three competing dashboards, none of which agree on what "down" means.
Platform Engineering: The team that builds the thing the other teams refuse to use.
Quantum-Ready: Currently works on a laptop, but with optimism.
RAG: Retrieval-Augmented Guessing. Now with citations.
Synergy: What two teams produce when they agree to disagree, slowly, in a recurring meeting Acornelius added to the calendar.
Tech Debt: Code Steve wrote three sprints ago. Now sacred. DELETE /technical-debt returns 202 Accepted and never completes.
Unblock: What a manager does after the ticket has been blocked for six weeks.
Velocity: The speed at which we pretend story points mean something. Ours is 47 per sprint. The historical average is also 47. Acornelius is suspicious.
Web3: The promise that the next time will be different. It will not be different. The pitch deck has been updated to say "AI" wherever it used to say "blockchain."
XaaS: Anything-as-a-Service. Yes, even regret.
YAML: Whitespace as a service.
Zero Trust: The realization that you should have been authenticating internal calls all along.

AI / ML Era Additions

Agentic: A workflow that used to fail deterministically and now fails with initiative. See also: Hallucination, Eval.
AI Governance: A quarterly meeting where legal discovers the product already shipped.
Anthropomorphize: What we do every time an LLM gets something right.
Copilot: Autocomplete with confidence and access to your insecurities.
Distillation: Making the model smaller, slower, and wrong in different ways.
Embedding: Coordinates for things. The map is also wrong. See also: Vector Database.
Evals: Unit tests for vibes. See also: RAG, Synthetic Data.
Fine-Tune: Memorize the eval set. Charge by the hour.
Foundation Model: The expensive black box underneath the inexpensive black box.
GPT-Wrapper: The startup category we are all in.
GPU-Poor: Has only one H100. Their startup is doomed.
Guardrails: Policy statements the demo ignores.
Hallucination Budget: The amount of money you're willing to set on fire before you admit the model is just making it up. See also: Hallucination.
Human-in-the-Loop: A person (Brent) manually fixing what leadership called automation.
Inference: The expensive part of being clever.
MCP (Model Context Protocol): A way for your AI to access more tools, creating more places to be wrong.
Multimodal: Wrong about more things at once.
Prompt Engineering: Writing instructions to a machine that ignores nuance like a junior product manager.
Prompt Injection: When the user asks more nicely than you did.
Reasoning Model: It thinks out loud. Some of the thoughts are correct.
Sycophancy: The model agrees with you. The model is wrong. So are you.
Synthetic Data: Fake data used to validate fake certainty.
System Prompt: The lie we tell the model so it will lie to the user.
Tokens: Money, but blurrier.
Vector Database: A place to store meaning until invoices arrive.
Vibe Coding: The fastest way to create tomorrow's archaeology.
Zero-Shot: We did not test this.

Traditional Additions

Backlog: A graveyard with priorities.
CI/CD: A machine that automates shame.
Data Lake: A swamp with a procurement process.
Feature Flag: A production if statement with a dashboard. See also: NutToggle (homepage).
Monorepo: One repo. Six engineers. Sixteen opinions.
Platform Team: People who built the golden path and are confused why everyone is in the woods.
Postmortem: A meeting where nothing is blamed and nothing is fixed. See also: our archive.
Tech Lead: The person asked to make architecture decisions after the architecture has shipped.

Update: RAG, Hallucination, and Generative AI (Sharpened)

RAG (Updated): Retrieval-Augmented Gaslighting. The model confidently retrieves the wrong document and then argues with you about it. See also: Hallucination, Eval.