All exercises
Act I·6 of 11·20 min
2.4

Tone-of-voice email

Write in your own tone of voice

Start a new Claude Code sessionClaude Code — start a new session.

We all know the term: "This sounds like AI" — we can either see it as a fault, or fix it by giving the agent your own voice. That voice lives in your <name>.md.

Prompt:

Draft me an email in my tone of voice using <name>.md

The message I want to transmit is:
The way companies are organized is about to change forever.
Not incrementally. Fundamentally.
We're moving from hierarchical org charts — divided by departments, full of approval bottlenecks — toward something completely different: intelligent, process-oriented organisations, where agentic workflows run bounded loops between human checkpoints.
This isn't a productivity improvement. It's a rethink of the entire structure.

Output to me directly so I can verify it here, do not make a real draft

Then ask yourself:

Is this mail something you would write?

Stuck?Debrief — what should have happened

A draft that sounds plausibly like you, not generic. Note the deliberate guardrail in the prompt: output to me directly, do not make a real draft. That's important — never let an agent send something on your behalf without you reading it first. Prompt injection from upstream content (the email you're replying to, the doc you're summarising) is a real attack surface. Verify before send.

Stuck?FAQ for this exercise

Q: How well does this generalise? Could I draft a Slack message, a memo, a LinkedIn post the same way? A: Yes — same pattern. The <name>.md is the voice; the prompt is the message. The trick is keeping the voice file rich (3–5 paragraphs of how you actually write).

Q: Is there a concern about confidentiality if Claude reads my emails? A: We're on an enterprise account so Anthropic doesn't save the data or use it for training — same trust level as Gmail or Dropbox holding your data. AI sounds scarier but it's the same kind of cloud database underneath.

OptionalGo deeper

Reverse the exercise. Download a colleague's person.md, then:

Read <colleague>.md. Draft an email FROM them TO me, in their voice,
summarising what they'd most want me to know about their week.

Output to me directly, do not make a real draft.

The same machinery in mirror — voice flows from the file in either direction. Useful when you're prepping for a 1:1 and want to imagine what the other side sounds like before they say it.


Ask for help