---
id: con-generative-ai
schema: concept/1
layer: garden
status: established
name: Generative AI
aka: ["GenAI", "AI chat assistants", "chatbots"]
summary: >
  Software that produces new text, images, audio or code in response to a request, by learning
  patterns from huge amounts of existing material. It drafts quickly, and you still judge the draft.
first_seen: "2022"
relations:
  builds_on: []
  evolved_from: []
  related: []
  parallels:
    - idea: "the printing press making copies cheap"
      field: "history"
      note: "When producing something becomes cheap, the scarce skill moves to choosing and judging what is worth producing."
implementations:
  - { provider: openai, name: "ChatGPT" }
  - { provider: anthropic, name: "Claude" }
  - { provider: google, name: "Gemini" }
  - { provider: microsoft, name: "Microsoft Copilot" }
  - { provider: perplexity, name: "Perplexity" }
  - { provider: xai, name: "Grok" }
competencies: [cmp-frame-the-outcome, cmp-use-responsibly]
origin:
  kind: ai-proposed
  by: "seed agent 2026-10-16"
  rationale: "Root of the map; every other card assumes the reader knows what these tools do."
---

Generative AI is the family of tools that make something new when you ask: a draft email, a summary of a long report, a picture, a slide outline, a few lines of code. They learned from enormous collections of human-made material, so they are good at producing things that look like what people usually write or make.

So what comes back is a plausible first draft. It can be excellent, generic or wrong, and it often looks equally confident in all three cases. Making a draft is now nearly free, so your time goes into deciding what you want and checking what you got.

## Example

A volunteer coordinator needs a thank-you note for 40 event helpers. She asks an AI tool for three versions in different tones, picks the warmest one, adds two real details from the day, and sends it. The tool saved her from starting with a blank page, and she added the judgment and the details.

## How it connects

- Most of today's text tools run on a **large language model**, the engine underneath.
- What you get depends heavily on **prompting** and the **context** you give it.
- Because drafts can be confidently wrong, **verification** belongs in every workflow.

## Try it

Pick something you need drafted and write one sentence describing what a good result would look like. Ask your AI tool for it, then hold its answer up against your sentence. Where did it match, and where did it drift?