AI is remarkable and we should all be using it reflexively to improve the quality of our work. Used well, it's a tool to improve our ideas and the quality of our work. It should not replace judgement, responsibility, or thinking. Here are some thoughts on how to use AI to support good work rather than fake it.
Part One: Ten Principles
1. Use AI for articulating, not authoring
Know what you want to say before you start an AI chat. Use AI to say it better, faster, or more clearly. But if you can't explain the purpose of your message without AI, the output will sound fluent but mean nothing.
Practical rule: Write your rough thoughts first, in your own words. Or record yourself talking through the idea and paste the transcript. Then read what AI gives you back and edit it by hand.
2. Keep the human trace
Good work contains things only you could bring. A judgement call. A constraint you chose. An idea shaped by your specific experience with this client, this project, this subject. Often small imperfections signal that someone actually cared.
When everything is polished to AI-perfection, it reads like no one took responsibility.
Practical rule: Treat every AI output as a first draft. Edit to preserve your voice, even if the AI version seems smoother.
3. Tone reflects your personality. Don't let AI set the vibe.
How you say something shows how you feel about who you're saying it to. When AI picks your tone, it's deciding your relationship for you.
Your clients and collaborators have a nose for AI-generated content. If the tone doesn't match what they expect from you, it lands as low-effort "slop" not worth engaging with.
Practical rule: Be specific about tone based on the relationship. "Make this direct because I'm accountable for the outcome" beats "make this professional." Give AI examples of how you actually write.
4. Don't outsource risk
Anything that carries personal, reputational, or relational stakes needs your judgement. AI removes friction. And friction is often the last checkpoint before harm.
Practical rule: If the message could upset someone, mislead them, or commit you to something significant, read it out loud and consider carefully before sending. That pause matters because the AI is not a throat you can choke when things go wrong.
5. Avoid pretending effort you did not put in.
Using AI to sound more thoughtful than you were is where trust gets eroded. People are remarkably good at detecting "faked" care.
Practical rule: If you didn't spend time thinking, don't send something that looks like you did.
6. Keep the source of the thinking straight in your own head
You don't need to disclose AI use constantly. But you need to know whether the thinking is yours or borrowed from AI. Confusing the two reduces your capacity for original thought, over time.
Practical rule: Ask yourself "Could I defend this position if asked why I wrote it?"
7. Use AI to compress, not inflate
Summarising your thoughts preserves intent. Expanding simple ideas into long text dilutes it.
Practical rule: Ask AI "make this shorter" or "more succinct" instead of "make this more impressive." Be the person who provides signal over noise.
8. Accept that less output can be more human
The temptation with AI is to say more because it's easy. Walls of text feel machine-generated. Brevity takes effort and people can feel that.
"I didn't have time to write a short letter, so I wrote a long one instead." - possibly Mark Twain
Practical rule: Delete paragraphs before you send. Tell your AI: "Be direct. Write tight. No preamble, no filler, no corporate voice. One idea per sentence. If a word can go, it goes."
9. Own your words
If you can't explain why you wrote it that way, you didn't write it. A roughly written message that you meant beats a polished one you didn't. One deliberate sentence is worth five generated ones.
Practical rule: Before sending, check that you can explain your choices.
10. Think of AI as an intellectual sparring partner, not a substitute
A sparring partner tests your ideas, exposes weak spots, forces you to defend your position. A substitute makes you unnecessary. Know which one you're using.
Practical rule: Did the AI challenge your thinking, or replace it? After using AI, is your thinking sharper or just... done?
Before sending, ask:
Could I explain in my own words what this is trying to do, why it's written this way, and why it's appropriate for this person?
If not, revise.
Remove the common AI tells
- Em dashes everywhere
- Bold words mid-sentence for emphasis
- The "It's not X, it's Y" rhetorical move
- Random emojis and decorative symbols
Part Two: Ten Ways AI Should Be Used
Here's where AI genuinely helps. These use AI to sharpen your thinking, not replace it.
1. Stress-test your reasoning
Use AI as a devil's advocate to find the holes in your logic before someone else does.
Prompt: "Here's my reasoning for recommending X. What are the weakest points? Where am I most likely wrong?"
2. Red team your proposals
Simulate how different people will push back before you present.
Prompt: "Act as a skeptical client, a budget-conscious CFO, and a technical lead. How would each critique this proposal?"
3. Break out of your first idea
Generate options when you're anchored on one solution.
Prompt: "I'm considering approaches A and B. Give me five alternatives I haven't thought of, including unconventional ones."
4. See through other people's eyes
Anticipate how different stakeholders will react.
Prompt: "How would this announcement land with: existing customers, prospective partners, and our development team? What concerns would each have?"
5. Map the downstream effects
Think through consequences that aren't obvious at first.
Prompt: "If we make this change, what happens next? And after that? Map the likely chain of effects over the next 3, 6, and 12 months."
6. Find your blind spots
Surface assumptions and missing information.
Prompt: "What would I need to know to be confident in this recommendation? What assumptions am I making that I haven't tested?"
7. Synthesise messy inputs
Turn scattered information into structured understanding.
Prompt: "Summarise the key tensions across this feedback. Where do people agree? Where do they fundamentally disagree?"
8. Draft to discover what you think
Generate a version specifically to react against.
Prompt: "Write a version of this recommendation. I'll use my disagreements with your draft to figure out what I actually want to say."
9. Make tradeoffs explicit
Force structured thinking about competing priorities.
Prompt: "What do we give up if we prioritise speed over quality here? Show me this as a tradeoff matrix with costs and benefits."
10. Rehearse hard conversations
Prepare for difficult discussions before they happen.
Prompt: "I need to tell a client their project is delayed. Play the client and push back on my explanation so I can refine how I'll handle it."