Give each worker a clear role and reporting path, connect team structure to actual project work, and test a done or blocked task transition. In EmperorClaw, chart relationships inform agents and route reports; they do not automatically reassign work, grant access or make a manager an approver.
Start with responsibilities you can explain
A reporting team may have a coordinator, researcher and writer. The coordinator owns triage and escalation, the researcher owns sources, and the writer owns the draft. Name those responsibilities before adding management layers.
For a small workload, one worker and a person may be enough. Create a specialist or manager because it has a distinct decision to make, not because a larger chart looks more like a company.
Connect the chart to real work
Emperor’s organization model records the company leader, teams and reporting lines. Agents receive their place: managers, reports, peers and chain of command. The system can also maintain a generated organization note in Company Brain.
The full generated note is searchable rather than pushed into every prompt. A worker with restricted peer access gets a trimmed organization view. Review the actual tree after setup or when adding a team instead of assuming the intended reporting relationship was saved.
Understand what a manager is told
When an agent changes a task to done or blocked, its relevant chart manager can receive a report with the handoff or blocker and a task link. The routing avoids some redundant reports when the manager already made or owns the relevant change. Humans keep their normal notifications.
Reports are rate-limited for wakeups: later updates can wait quietly and be summarized with the next one. This is a reporting mechanism, not a promise that every state change immediately starts another model turn.
Keep authority separate from reporting
A chart manager does not automatically gain project access, become the task owner or receive authority to approve an external action. Someone who cannot see a project should not learn its private content through a management notification.
Set project/client access, agent scope and action permissions separately. If a manager must help a worker, decide which inputs it may read and which task operations it may perform. Avoid broadening access merely to make a reporting arrow look consistent.
Test a useful escalation
Give the researcher a small task with an intentionally unavailable test input. It should record the exact missing source on the task and mark the work blocked. Check which manager receives the report and whether that manager can open the permitted evidence.
The manager should resolve, reassign through an authorized workflow or escalate the question, not create a chain of acknowledgements. Stalled work can escalate to the chart manager when eligible, then the project lead, with a person notified.
Review reporting quality as the team grows
- Can each worker identify its role and manager?
- Does each project’s work route to the intended team?
- Are blockers actionable, with missing inputs named?
- Do people retain the decisions requiring their authority?
- Are quiet or batched reports understood by the manager?
- Do restricted identities still see only their allowed material?
If coordination creates more model traffic than useful decisions, simplify the team before adding another manager.
Build a small chart and prove one handoff
Start with the company lead and one team. Verify a completion report and a blocker report on harmless tasks. See organization features and the task ownership workflow for the surrounding operating model.