The container moment

Risk professionals know the black swan: huge impact, impossible to predict. Most of what hits supply chains today is something else. It is the grey rhino, the big, visible threat that everyone sees coming and too few prepare for.

When Washington started to move on tariffs, many companies discovered they did not know exactly who their suppliers were. Everyone got busy mapping. That is the rhino in action: visible, foreseeable, and still a surprise to the organisation.

Take the first two forces as axes and four worlds appear. Select one.

In the 1950s, Malcom McLean put cargo into a steel box with two doors. You open it, you put things in, you close it. Nobody at that desk thought he was rewriting world trade.

Yet the box standardised everything around it. Ships, cranes, trains and trucks were rebuilt to one format, and loading stopped being a matter of bags, drums and small cases. The cost of moving goods across oceans collapsed and global trade took off.

The part that matters for us: the value did not arrive with the box. It arrived over the following two decades, as ports, labour agreements, rail networks and business models reorganised around it. Ports that adapted became hubs. Ports that did not, faded.

The bottleneck is organisational, not technological. That is the container lesson for agentic AI.

Figures for McLean's first container voyage on the Ideal X, as reported in Marc Levinson, The Box. A 97% drop, from one box.

Who uses Excel every day? In the room, nearly every hand went up.

In a McKinsey survey on planning tools, around 80% named Excel as their main planning system. A pharmaceutical company with 20 billion revenue had just finished implementing a new planning system. We found seven people using it regularly. Everyone else was in Excel.

Searching, combining and reporting data is waste in the lean sense. We do it because no system does it for us. When AI takes over that assembly work, the risk manager's week flips from war rooms and manual reporting to early warning, scenarios and diversification.

In workshops, people often treat AI as something magical that solves anything without context. It does not.

Think of a new colleague straight from university. Very smart, no idea how your business works. Hand over a task without explanation and they will not know what to do. The model is the same. You onboard it.

1. Onboarding. Share background, assumptions, market dynamics, customer priorities, constraints.
2. Mentor role, the "Meister". An experienced planner explains cause and effect and what matters most.
3. Reverse mentoring. The digital colleague challenges assumptions and proposes alternatives.
4. First tasks, interaction and feedback. Real scenarios, corrections, refinement.
5. Standardisation and documentation. What worked becomes the way you work.

The demand planner asks the system for the forecast three months out. It says 517. "Why 517? I expected 550." No answer. So the planner rebuilds it in Excel, because there at least they know what they did.

Put a GenAI layer between the planner and the backbone. It translates the question into data queries and models, and explains what comes back. You talk to your system the way you would talk to a colleague. That vision is decades old. Now it works.

I wanted to understand it myself, so I built a sample company (suppliers, customers, channels, two years of P&L, no confidential data) and vibe coded demand planning, supply planning, risk and resilience on top of it.

• Full planning functionality, natural language input and scenario calculation in a clean interface.
• Deep business knowledge is the crucial input. Thirty years of experience let me challenge every step.
• Still an MVP. Industrial scaling needs roughly another 50 hours.
• The surprise: in the multi-agent simulation, trust between functions rises, then falls again if incentives stay unchanged. Agents behave like people.