Glossary

Buzzwords, explained with analogies from the trades.

We use them because they name real things. Each one comes with an analogy and a critical eye: what it is, when it helps and when it does not.

RAG

A well-organized archive with a librarian who knows where everything is and cites the source.

Retrieval-augmented generation: before answering, the AI searches your documents and builds the answer from what it finds.

When it helps: When you have a lot of in-house documentation and people lose time looking for things.
When it does not: When the documentation is disorganized or out of date. The archive has to be put in order first.

A new employee with written instructions and a supervisor who checks the work.

An AI that does more than answer: it carries out steps. It searches, fills in, sends and records.

When it helps: For repetitive tasks with clear rules and a human checkpoint.
When it does not: When the task requires judgment nobody can write down, or when nobody will review the work.

The standard plug that lets one tool connect to another without adapters.

A protocol that lets AI assistants connect to your systems in an orderly way.

When it helps: When you want AI to read from or write to your tools without custom development for each one.
When it does not: When the real problem is that the data does not exist or was entered incorrectly.

The technician who comes to your home instead of selling you the appliance from a catalog.

Forward Deployed Engineer: someone who works inside the client's operation so the technology actually works.

When it helps: When the problem is not the tool, but how it fits the way you work.
When it does not: When the problem is small and a standard tool already solves it.

The plumbing: if it is badly designed, it just moves dirty water faster.

Making something happen on its own, without a person triggering it every time.

When it helps: For frequent tasks with clear rules and reliable data.
When it does not: When the process is not defined yet. Organize first, then automate.

The breaker panel: it decides what can flow and trips when something fails.

The rules on who uses AI, with which data, for what purpose and under what review.

The building's blueprints: without them, every repair starts from scratch.

The company's information, organized by department, process, policy and form.

When it helps: When knowledge lives in the heads of a few people, or walks out the door when someone leaves.
When it does not: When nobody will maintain it. An outdated knowledge base is worse than none at all.

The work order: the clearer it is, the better the job turns out.

The instruction you give the AI, with context, a goal and constraints.

When it helps: Always. A good prompt changes the result more than switching tools does.
When it does not: When it is used to paper over a poorly designed process.

The employee who would rather make something up than say "I don't know."

When the AI states something false with complete confidence.

When it helps: Never. But it can be kept in check with cited sources, human review and acceptance tests.
When it does not: It cannot be eliminated entirely. That is why every managed task has a checkpoint.

Sending someone on a specialist course: it does little good if they do not have the blueprints first.

Retraining a model on your own data so it behaves in a particular way.

When it helps: In specific cases, with many high-quality examples and a very particular need.
When it does not: For most small and mid-sized businesses: solid RAG and good prompts get there sooner and for less.

A new term, now and then

When a buzzword worth knowing comes along, we will explain it to you with an analogy.

Occasional updates, resources and ideas to keep learning.

In a Speedboost we look at them in the context of your business, live, in two hours.