I wrote up the overall pattern in Running a Team of AI Bots With Grok Bot. This one zooms in on a single piece: the research bot, and how a daily research run is set up in this kind of team.
The job
One bot, one job. The research bot doesn’t write posts. It scans news, changelogs and release pages for the beats you care about and hands back short briefs, each with a source link.
How the daily run works
- A schedule. A routine kicks it off once a day, so research is done before drafting starts.
- A freshness window. Only stories from roughly the last day or two. No fresh source, no brief.
- Beats with a floor. You can tell it to always include a minimum share of a topic so one beat doesn’t crowd out the rest.
- Dedupe first. Before anything gets drafted, briefs get checked against what’s already on the site, drafts included. Repeats are skipped, and real follow-ups become clearly labeled updates.
- Hand-off, not publishing. Briefs go to the editor bot, which writes drafts only. A human decides what goes live.
Why split it out
Keeping research separate means the editor bot isn’t burning tokens browsing, and the research bot can’t accidentally publish anything. Each bot gets a short rule list, and each reports back to whoever asked.
My take
The rule I’d start with: do the job by hand first, save it as a skill, and only automate what has worked twice. Research is a great first job to hand off because it’s easy to check.
