Using generative AI at work
Generative AI can save hours each week — if you use it on the right tasks, ask clearly and stay inside your organisation's rules.
Learning objectives
- Identify good, low-risk use cases for generative AI
- Use the 4D Framework and three delegation questions to decide what to hand to AI
- Write a clear prompt using task, context, format and constraints
- Explain why approved tools matter and what 'shadow AI' is
3.1 Where generative AI helps most
Generative AI is most useful for tasks where a good first draft saves time and you can judge quality yourself. Common, low-risk starting points:
- Drafting and rewriting — emails, reports, job adverts, policies, social posts
- Summarising — long documents, meeting notes, research papers, email threads
- Thinking partner — brainstorming, outlining, stress-testing an argument, role-playing a difficult conversation
- Explaining — a new topic, jargon, a spreadsheet formula, a piece of code
- Transforming — tone, length, reading level, format (notes → table), translation
- Analysis support — exploring data, suggesting charts, spotting themes in survey comments
3.2 Deciding what to delegate
Anthropic's AI Fluency course, developed with academics Joseph Feller and Rick Dakan, organises good AI use around four competencies — the 4D Framework:
| Competency | The question it answers |
|---|---|
| Delegation | Should AI do this at all — and which parts should stay with me? |
| Description | How do I clearly explain what I want, with the right context? |
| Discernment | Is the output accurate, appropriate and good enough? |
| Diligence | Am I using AI responsibly, transparently and in line with my obligations? |
For delegation, ask three quick questions:
- Can I check the result? If you can't judge whether the output is right, AI is risky for that task.
- What happens if it's wrong? A clumsy brainstorm costs nothing; a wrong figure in a board paper or a wrong answer to a customer can cost a lot.
- Is the data appropriate to share? We cover this in Module 4.
3.3 Asking well: a simple prompt pattern
Treat the AI like a capable new colleague who knows nothing about your situation. Tell it the task, the context, the format you want, and any constraints.
Vague
"Write an email about the office move."
Clear
"Write a short, friendly email to 40 staff announcing that our Ipswich office moves to the second floor on 3 November. Include: desks must be cleared by 31 October; IT will move computers; the new floor plan is attached. Under 150 words, UK English, no exclamation marks."
Then iterate: ask for changes ("shorter", "more formal", "add a line about parking") rather than starting over. The Prompt Engineering course takes this much further.
3.4 Approved tools and your organisation's policy
Before using any AI tool for work, find out:
- Which tools are approved — often an enterprise version of ChatGPT, Claude, Gemini or Copilot — and which are not.
- What data you may put into them (Module 4).
- Whether you need to disclose AI use — for example in client work, publications or recruitment.
- Who to ask when you're unsure.
Using personal, unapproved AI accounts for work is common — and it's how confidential data most often leaks. If the approved tools don't meet your needs, raise it; don't route around it.
List three recurring tasks in your week. For each, run the delegation questions: can you check the result, what's the cost of an error, and is the data safe to share? Pick the one with the best answers and try it with an approved tool this week.
Key takeaways
- Start with tasks where AI drafts and you can judge the quality.
- Delegation, Description, Discernment and Diligence are the four AI fluency competencies.
- Clear prompts state the task, context, format and constraints — then iterate.
- Use approved tools; unapproved personal accounts are a common source of data leaks.