AI at Work

This hub covers the practical side of working with AI assistants such as ChatGPT, Claude, Gemini, or Microsoft Copilot: reusable prompts, checking output, team ground rules, and learning with AI. It's for people who want results they can depend on: a developer asking for a migration script, a student turning lecture notes into practice questions, a team lead asked to draft the team's AI guidelines.

Getting an answer from AI is easy; getting a good one on purpose, getting it again next week, and knowing when to trust it is harder. The failures are mundane: the prompt that worked lives in one person's chat history, a wrong figure reaches a client inside a confident paragraph, rules get improvised one awkward question at a time, and a topic feels learned because the summary read smoothly. The guides below tackle each of these with templates, checklists, or decision rules you can use this week.

Save prompts as templates, not chat history

A prompt that worked once is only worth keeping if someone can reuse it without rewriting it, which means a template: fixed instructions plus named slots, such as {{audience}}, for the parts that change. Our guide to writing reusable prompt templates shows what a reliable one contains, with examples to copy. Shared templates also need upkeep: building a team prompt library covers owners, last-tested dates, and pruning stale entries before they erode trust in the rest.

Match the checking to the stakes

AI output sounds equally confident whether it is right or wrong. Models can invent sources and figures (usually called hallucination) or describe settings menus that have changed since their training data was collected. Checking everything equally wastes the time AI saved, so ask what a mistake here would cost: a brainstorm needs a glance, a figure headed for a customer needs tracing to its source. The guide to checking AI output before you trust it turns that into a review checklist for each level of risk.

Write the ground rules down

A team that uses AI tools already has an AI policy; it's just unwritten, and everyone holds a different version. One page settles most of it: approved tools, information that never goes into them, outputs that need a second reviewer, and when to disclose AI use. In the EU, the AI Act's AI-literacy duty has applied since February 2025, and the European Commission's Q&A says it covers a company whose staff use ChatGPT to write or translate text: they should be told about risks such as hallucination.

Keep the thinking that's meant to be yours

A clear AI explanation is easy to mistake for your own understanding; the test is whether you could reproduce it with the chat window closed. Patterns that keep you thinking make the model a questioner or critic: it quizzes you, finds the gaps in an explanation you wrote first, or gives a hint instead of the answer. Learning with AI without fooling yourself covers patterns like these. When AI reshapes training material, the guide to turning documentation into audio explains why audio suits the "why" and "what changed" while the written document stays the record.

How to use this hub

Read these three guides first, in order:

1. Reusable prompt templates, the skill the rest of the hub builds on.
2. A prompt library for your team, useful even if you work alone, for its last-tested dates.
3. Learning with AI without fooling yourself, before you use AI on anything you need to understand, not just produce.

Then add the review checklist from the guide on checking AI output, and the policy template once AI-assisted work passes between colleagues. If training material is your main problem, start with the docs-to-audio guide.

Templates work in any text file. DocPod, our experimental agent that turned policy documents into two-host audio episodes, is offline while we rework it, and the audio guide's method works without it.