Notes for owners · July 2026

AI for client communication: where it shines, and where it shouldn’t go alone.

Client communication is where AI pays best and where it can hurt most. It’s recurring, it’s voice-dependent, and it eats owner hours: a perfect drafting target. It’s also read by the people who pay you, in moments that sometimes matter enormously. Both things are true, and a serious owner plans for both.

Where it shines

The recurring middle of client communication is drafting work, and a trained system does it at your standard: the monthly report, the status update, the meeting recap sent same-day, the answers to questions you’ve typed a hundred times, the follow-up while the conversation is still warm. Speed matters here: the same-day recap reads as care; the Thursday version reads as afterthought.

Where it shouldn’t go alone

Some messages are decisions wearing an email’s clothes. Delivering hard news. Anything that changes money, scope, or expectations. The message after a mistake. In a law practice, anything that brushes against advice; in family-building work, the sensitive milestones where one wrong word does real damage. A governed system still helps here, drafting from your written rules for tone and content, but the human doesn’t just review these. The human decides.

The review architecture that makes it safe

• Written rules first: what the system may say, must never say, and can never invent, approved before it drafts a word
• Tiered review: routine drafts get a read; consequential ones get the owner’s judgment
• Nothing sends itself: the system drafts, a person sends. That order never reverses

Owners sometimes read this as caution and expect the drafts to be timid. It’s the opposite: because the system can’t invent and can’t send, it can be trained to write with your full directness, the awkward emails included, firm and graceful, while the risk stays exactly where you put it.