Skip to content
BENIBIX Learn

Working with AI

Prompting, instructions and agentic coding · 35 terms

Sort
Compare

Prompts

9

What you send to a model, which model gets it, and how you write and test it.

The text you give an AI model to tell it what you want.

A chat with many messages back and forth. The model rereads the whole chat each turn.

You write a prompt, test it and fix it. You stop when the answers are good.

Rules set before the chat starts. They shape each answer the model gives.

Notes you save once. The app adds them to every chat for you.

A prompt with blanks that a program fills in each time it runs.

A shared set of tested prompts that staff can copy and reuse for common tasks.

Picking what a model sees for each task, and in what order.

Picking the AI model that fits a job. You test a few first.

Prompt Writing Techniques

11

Ways to write a prompt that get better answers. They go from worked examples to step-by-step thinking.

Prompts with Examples

3

Prompts sorted by how many worked examples they show the model, from none to several.

You ask a model to do a task. You give it no examples.

A prompt with one worked example in it. The model copies its shape.

You put a few solved examples in the prompt. The model copies their pattern.

Prompts for Reasoning

3

Prompts that make a model work a hard problem out in steps, paths or branches before it answers.

You ask a model to show its steps first, then give the answer.

Ask the same question many times. Keep the answer most runs agree on.

The model tries a few next steps. It keeps the good ones and backs out of dead ends.

Persona and Format

2

Ways to shape one prompt: who the model should be, and where each part starts and ends.

You tell a model who to act as, like a tutor or a coach.

A mark that fences off one part of a prompt from another.

Multi-Step Prompts

3

Getting a result over several prompts: a chain of steps, rounds of changes, or a prompt a model wrote.

Splitting a task into steps, where each prompt feeds the next.

You ask, read the answer, then ask for changes. You repeat until it is right.

Asking a model to write or improve a prompt for you.

Output Control

4

Ways to make a model's answer fit a set shape. Also ways to make repeat requests cheaper.

The model's answer must fit a set shape, like JSON with set fields.

A switch so the model only replies with JSON that will load.

A button that asks for a fresh answer to the same message.

Saving the work on a prompt's fixed start, so the next request reuses it.

AI-Assisted Coding

11

Writing software with AI, from hints as you type to agents that run the tests.

An AI writes program code from a plain request.

AI that suggests the next lines of code as you type.

An AI that codes on its own. It edits files and runs the tests.

Making software with AI and not reading the code it writes.

Write what to build first. The AI plans and codes from that written spec.

A free method that gives AI agents team roles. They plan first, then code.

A free GitHub toolkit. It walks an AI agent from a written spec to code.

An AI coding tool from AWS. It writes requirements, a design and tasks before the code.

A file with the house rules for AI agents in your code.

An AI reads a code change and flags bugs and risks. A person decides.

Future cleanup work caused by AI code that was added without care.