Sultan
Sultan Operating Layer
Humans still own legal authority. Sultan exists to route governed agent work, preserve auditability, and keep execution inside approval law.
Active agents
Governed agent roster
47
Tasks waiting
Drafts, reviews, and escalations
3
Live sessions
Human + agent workspaces
1
Memory entries
Expiring working memory
48
Next Moves
Approval pressure, policy blocks, dead letters, and dry-run readiness stay visible before autonomous behavior grows teeth.
Approvals and Blocks
Where governed AI work waits on a human gate or policy repair.
Draft delay apology
Draft a policy-safe response for a delayed white-glove delivery with refund pressure.
Draft first-touch sequence batch
Generate 24 personalized draft emails for Milan and Dubai hospitality leads.
Prepare RXO packet
Complete missing service level and dimensions before RXO delivery packet can move.
Performance Drift
Agents whose quality or confidence needs tuning attention.
Chief of Staff Agent needs tuning
27% human correction rate, 1 policy blocks.
Knowledge Agent needs tuning
27% human correction rate, 2 policy blocks.
Memory Agent needs tuning
18% human correction rate, 3 policy blocks.
Evaluation Agent needs tuning
27% human correction rate, 0 policy blocks.
Governance Agent needs tuning
27% human correction rate, 1 policy blocks.
Planning Agent needs tuning
27% human correction rate, 3 policy blocks.
Sultan Governance
Sultan should feel like an accountable operating layer: scoped, bounded, owned, and visibly governed.
Risks
Next Actions