Yesterday a client suddenly needed another version. I was editing images while thinking: the package hasn't shipped, tomorrow's meeting isn't booked, next week's flight isn't reserved. I'm doing three people's jobs alone.

They split the work across 14 "bot employees"

Peng Zheng (Peng) is the lead designer at an overseas AI company's Grok project. Lauren is an engineer there. Their team has not just colleagues but a bunch of dedicated AI bots, each handling its own lane.

The three that stuck with me:

First is Peng's "chief butler bot" — it orders his office supplies and tidies up the equipment list. He says: "I save 3 hours a week just thinking clearly about what new things I need."

Second is Lauren's "travel bot." Last time she asked it to plan a trip across three cities. The bot compared prices, picked times, and produced a plan — Lauren only did one thing: approve.

Third, my favorite, is Dr. Eggbot. A bot whose only job is auditing your other bots. AI writing isn't the end — it does another pass. Peng says: "I used to skip review. Now I'm forced to do it."

You don't need 14 bots — just learn three moves

Core idea in one line: swap "I do one whole lane alone" for "I do the first 5%, then let AI scale the rest." Peng said: "I'll do the first 5%, lock in the key parts, then let the bots generate the rest."

Apply it to your work: before writing a Xiaohongshu post, tell AI "this time emphasize XX feeling," let it draft 10 versions and you pick; before booking a trip, tell it the time, city, and budget, and let it draft a plan first.

Replication cost:

  • Money: $0 — ChatGPT, Kimi, Wenxin Yiyan all work
  • Time: 30 minutes today — pick last week's most painful task and ask AI "can you help me handle this in X steps"
  • Tech barrier: if you can type, you're good — no code needed, totally fine
  • First step: open your usual AI tool and say "from today you're my [travel butler / content assistant], I'll give you a task every Monday"

How three types of people should start

If you're just starting out / still testing direction: pick 1 most repetitive task (like replying to client template messages) and hand it to AI. You'll find you save 5 hours a week — and those 5 hours are what you should spend finding direction.

If you already have 1-2 stable clients: pick client comms and invoice tracking, and hold a weekly "meeting" with AI. You don't need many bots — only ones that work reliably actually matter. My past mistake: I rolled out 8 scenarios at once, all crashed.

If you're already scaling: learn Peng's line "let AI audit AI." You use AI to write things, right? After writing, copy it to another AI role whose only job is picking flaws. I used to publish after one read — I got stuck several times myself. Bots critiquing bots can patch those holes.

Final thought: not everyone needs this approach. If you don't have many clients yet and your work isn't very repetitive, focus on doing the work well first — bots are for the busy, not the idle.