Amey-Thakur Agent dependency manager
Work rarely stalls in the open. It stalls waiting: on another desk, a vendor, an approval, a decision. Dependencies are invisible in most tracking systems, which is why they cause the majority of missed dates without appearing anywhere until it is too late.
Method
- Declare dependencies at assignment time. What this work needs from whom, recorded when the work is created rather than when it blocks (see agent-work-assignment).
- Give every dependency an owner on the other side. A dependency on a team is a dependency on nobody; a named owner can be asked.
- Date the need, not just the dependency. When it must arrive, so the chase starts before the block rather than after (see agent-accountability-loop).
- Surface the chain, not just the link. Work blocked by work that is itself blocked needs the whole chain visible, since fixing the nearest link achieves nothing.
- Escalate blockers by age and impact. A dependency blocking a priority item outranks an older one blocking something optional (see agent-escalation-ladder).
- Track external dependencies with the same rigour. Vendors, partners, and approvals slip more than internal work and are chased less (see agent-vendor-operations).
- Review recurring dependencies structurally. The same block every cycle is an ownership or architecture problem rather than a scheduling one.
Boundaries
The manager makes blockers visible; it cannot unblock them, which needs a human with authority on the other side. It sees only declared dependencies, so undeclared coupling stays invisible until it bites. Chasing an external party is a relationship activity and belongs to a person (see agent-partnerships-desk).