A daily loop: a metrics Bot decides, other Bots change the s

How this helps you

Your content keeps improving from real data without you running each step.

✦ Try it

Write a skill called [loop name]: ask [metrics Bot] for the last 24 hours of [metric] per [item], pick the bottom [3] (if tied, pick at random so it's exactly [3]), ask [designer Bot] for [3] replacements, ask [builder Bot] to swap them in and deploy, then send me a short brief of what changed.

Fill in: [metric] and [item] are what you measure (page views per sticker). The Bot names are your own.

What they did

In Grok Bot 201, Joseph from SpaceXAI's field engineering team shows the step that took his sticker shop to full autonomy: a fourth Bot, Ed, for metrics. His primary Bot, Roy, runs a sticker PMF skill every 24 hours. Roy asks Ed for page views from Vercel analytics (Ed signs in to Vercel on its own computer; he says an API token also works and secrets are kept separate from the Bot's computer). The bottom three stickers are cut, Dex (a designer team Bot) makes three new ones, and Lou (a full stack Bot connected to GitHub) swaps them in and deploys. He added a tie rule after an edge case: pick at random so exactly three go. He says he spent about 30 minutes with each Bot getting its part right before joining them up, and his co-host suggests a daily brief as a human checkpoint.

Who it's for: People with a site, shop or content feed and clear numbers to judge it by.

Before you start

A site or product with analytics a Bot can read.
Specialist Bots that already do their part well by hand.
A primary Bot to orchestrate.

How to try it

Get each Bot's single step working on its own first.
Give a metrics Bot access to your analytics.
Ask your primary Bot to write one skill for the whole loop, naming each Bot's step.
Add rules for edge cases such as ties.
Run it by hand and ask for a diagram of the flow.
Then schedule it every 24 hours, with a daily brief to you.

What to expect

A daily run that messages each Bot in turn and ends with a brief: what was cut, what was added, and the deploy.

Common mistakes

Joining Bots up before each works alone. He tuned each one for about 30 minutes first.
No checkpoint. Keep a daily brief so you see every change.

Related tips

Pick a main Bot so it can see and delegate across your team: A main Bot gets wider visibility into your other Bots, messages you first, and hands work out.
The fan-out method: three models, one brief: Launch Cursor cloud agents on three models at once and keep the best parts.
How Pages stay up to date (and what they cannot do): Pages are not live: a routine rewrites them. Know the limits before you build many.