How to make AI agents work together

The chief of staff is the only bot that talks to you about priorities

Every request you make, every routine that fires, and every ticket that arrives goes to one bot first. It creates a task file with an owner and a line that says when the task is done, assigns it, and posts it in the pod room. It never sends email, publishes, or changes an external system itself; it routes and it reports. That single rule removes most of the confusion in a multi-agent setup, because there is exactly one place to ask “what is happening” and one bot whose job is to know.

It also enforces a work-in-progress limit of three tasks per bot, flags anything stuck in Doing for more than two days, and moves anything waiting on you into a list you see twice a day.

Pods, and why the group chat size matters

Organize bots by function into pods of four or five, each with its own group chat that also contains the chief of staff and the reviewer. On Grok Bot a group chat holds up to six bots, so a pod of four plus those two fills a room exactly. A brand pod, an operations pod, a CRM pod, an SEO pod, an ads pod, a local-and-social pod, a sales pod, a build-and-content pod: eight rooms, each with a clear charter, is easier to follow than one room with thirty voices.

Handoffs are explicit: the sending bot mentions the receiving bot with the task ID in the pod room; the receiver acknowledges in its next turn or the chief of staff reassigns. Status updates use one shape: task ID, status, next step.

The task file and the board

One file per task, in a shared folder every bot can read. The template is short: title, owner, who requested it, which company it is for, the goal in one sentence, a done-when line you can check without opinion, the access tier, approvals needed, inputs, deadline, a timestamped log, and the lesson line at close. The board has six columns: Backlog, Ready, Doing, Waiting on human, Review, Done.

Definition of done is written per task type. Content is done when it meets the quality bar with every fact sourced. A report is done when every number traces to a saved screenshot. Outreach is done when the packet was approved, sent, and the reply is tracked. A build is done when the QA checklist passed and a human opened the staging link.

The approval packet: how a human decides in seconds

Nothing reaches you as a raw chat message. Every action that touches the world is submitted in one shape:

APPROVAL · bot name · task ID · tier
WHAT: one sentence, plain words
TARGET: exact recipient, URL, record, or campaign
WHY: the trigger and the goal
SOURCE: link to the email, ticket, or instruction that started this
REVERSIBLE: yes / partly / no, and how
PREVIEW: the full draft, screenshot, or exact command
CHECKS: facts verified, quality items met, who else it affects

One packet is one action; no “and also.” If any field is unknown the packet is not ready. The reviewer reads it first and returns pass or fail with the rule it applied; fails go back to the owner. Passes go into the chief of staff’s digest at 8:00 and 16:00, ranked money and security first, then client-waiting, then everything else. Urgent means a client is waiting or money or security is involved, and those come immediately. On Grok Bot you answer Allow once, Deny, or Always allow from the desktop or the phone (approvals, security and privacy).

One shared root: the brand context

Coordination fails when two bots hold two different versions of a fact. The cure is one document per company that every bot reads and only one bot edits: identity and locations, canonical services, voice with say and never-say lists, market and competitors, digital footprint with every ID, rules of engagement, assets, plan, change log. Its status gates the work: nothing is built, written, posted or advertised until it is approved, and every deliverable cites the version it used. A consumer that needs a fact not in the file asks the owner to add it; it does not improvise.

Rhythm: standup, digest, review

• 7:30 standup, written by the chief of staff: done yesterday, doing today, blocked, approvals waiting, risks.
• 8:00 and 16:00 approval digests.
• Friday 15:00 review with the learning officer: throughput, approval rate, denials, incidents, lessons promoted.

Every one of these is a routine on the chief-of-staff bot, created after the workspace is seeded. The mechanics of creating a bot and its routines are in How to set up a Grok Bot agent; the reason the team is structured this way rather than as one assistant is in Why one AI agent is not enough. Multi-agent coordination is an active research area; the OWASP GenAI Top 10 also lists excessive agency as a risk, which is exactly what the reviewer and the packet format are there to limit.