How a mid-market fashion manufacturer rebuilt its logistics function around an AI Worker that owns the full container lifecycle — and cut manual coordination by more than 50% in a few months.
Shuya Zhai, Logistics Director at Lamour, and Philippe Marcotte walk through how an AI Worker now runs container logistics end to end — and where the team's time goes now.
Lamour's logistics team was managing thousands of inbound container shipments a year across dozens of overseas suppliers, multiple freight forwarders, and several ports of entry. The work itself wasn't complex — but the volume of coordination, document chasing, and reconciliation was consuming the team's entire week.
The team wasn't short on talent.
They were short on capacity.
Every hour spent chasing a supplier or reconciling a packing list was an hour not spent on the work that actually required judgment.
Before building anything, we ran a full process diagnostic of Lamour's logistics department. Every handoff, every system, every document, every escalation path — mapped end to end.
Only then did we draw the line between work that didn't require a human, and work that did. The AI Worker covers the first. Claude and Claude Cowork support the second.
This is the difference between deploying technology and modernizing operations.
The AI Worker doesn't automate one step. It owns a responsibility — the inbound container lifecycle — from the moment a cargo-ready window opens 28 days out, to the moment the container clears customs and lands in the warehouse.
The reconciliation logic is where the AI Worker earns its place in the org chart. Three checkpoints, three different decision rules — each one explicit, auditable, and reviewable.
The AI Worker cross-checks the Commercial Invoice, Packing List, and Bill of Lading for internal consistency. If a SKU appears on the invoice but not the packing list — or vice versa — the supplier is asked to correct and resubmit. The AI Worker handles this loop without involving the team.
Once the documents agree with each other, they're compared to what was actually ordered in the ERP. Quantities are matched line by line. Attributes — color, size, model — are matched exactly.
If discrepancies in quantity stay under 5% and all attributes match, the AI Worker proceeds. If discrepancies exceed 5%, or if an attribute doesn't match, the order is escalated to a human reviewer — with full context, the source documents, and a one-click decision interface.
The point is not that the AI Worker is always right. The point is that it knows the difference between "I can handle this" and "a human needs to look at this" — and it routes accordingly.
The AI Worker took the deterministic half of logistics. The team kept the judgment half — and got better tools to do it. Claude and Claude Cowork are now where the logistics team works on everything that doesn't fit a rule.
We built a custom MCP server that exposes Lamour's ERP directly to Claude — and we deliberately scoped it to read-only access. The team can query the ERP, run reconciliations, summarize containers in transit, and pull reports in natural language without leaving the chat interface. But nothing in the ERP can be modified through this channel. That was the level of access everyone — Lamour, IT, and SNP — was comfortable with. It's the right default: highly useful for the team, and safe by construction. ERP writes stay with the AI Worker, where every change is governed by explicit business rules and full audit logs.
The AI Worker reached production inside the logistics department in a matter of months. The results compounded as the team's time shifted from running the process to improving it.
No noise. No product updates. Just Philippe's read on where work is heading and what it means for how organizations need to operate.
Bi-weekly. Unsubscribe anytime.
