The email my AI wrote last week almost cost me a client
Xiaolin runs a Xiaohongshu consulting client I work with. She vented to me last week: she had AI write a reply to a client, the tone came off too aggressive, and a long-time client canceled next month's service on the spot. I've made this exact same mistake — I once let AI auto-reply for me and ended up quoting the client at 2x the right price.
Right after we finished talking, the Utah news popped up on my feed.
Utah let AI write acne prescriptions — the real fight is "who's on the hook"
Utah just approved a company called Nolla Health to let AI write initial acne prescriptions for patients. Note: not assisting doctors — the AI writes them, doctors check at stages.
The moment this dropped, the debate exploded. If it prescribes wrong, who's responsible? The doctor says "AI wrote it," AI says "the doctor signed off," and the patient is stuck in the middle.
Same problem we have with AI writing contracts, replying to clients, making decisions.
I made this exact mistake before: let AI send a client quote without my eyes on it. Ended up writing 80000 instead of 8000, the client screenshotted it and posted it in a group. We didn't close the deal, and my brand reputation took a hit.
This news reminded me: AI is getting more "automatic," but "automatic" ≠ "off the hook."
Replication cost: $0 + 1 hour + no tech skills
- Money: $0 (the news itself is free)
- Time: 1-2 hours (map out which parts of your workflow use AI)
- Tech barrier: no code needed, just open your usual AI tool
- First step: open whatever AI tool you used recently (Doubao / Kimi / ChatGPT all work), scroll through what you let it "automatically" handle last week. Ask yourself: if this step went wrong, would I know immediately?
Concrete actions:
1. Important external content (contracts, quotes, complaint replies) must have a human final check
2. Set a "red line list" for AI — what types of decisions it can't make alone
3. Have an "AI went wrong" playbook — how to tell clients, how to recover
This isn't urgent, but it's coming. I only started writing my own list last month.
Advice by stage
If you're just starting out (still finding clients): I'd treat AI like an "intern" — important stuff, always run it through yourself. No rush to test it now, but build the habit.
If you already have 1-2 stable clients: I'd spend a weekend listing all the "AI-automated steps" in your workflow, mark which ones are high-risk. I only did this after that quote disaster myself.
If you're scaling (about to hire or already have a 1-5 person team): I'd strongly recommend writing up an "AI Usage Policy" — even just one page, spelling out what AI can decide alone vs. what needs human review. Not complicated, but it'll save you when something goes wrong.