Somewhere in most people’s houses, there is a road atlas. Beautifully printed, comprehensive, the place you were looking for always on the join between two pages or on the fold where the binding had eaten the postcode. Then the internet arrived and you printed out a route from a website, which was a digital map in name and a dead letter on the passenger seat. The thing that actually changed driving was location services. The moment the map knew where you were and where the closures were, it stopped being an artefact and became a network you could query. Then Waze went further: routing you using information contributed by every other driver on the road. A car on the hard shoulder. A speed camera. A tailback the official feed would not catch for twenty minutes. The map was no longer a thing you read. It was a network where every participant made the whole more intelligent for everyone else.

So, time to drag this back to my actual day job, which is business networks and S/4HANA. Every S/4 programme has to answer a deceptively simple question: how will this enterprise communicate with the outside world once the transformation is done? The old plumbing breaks. IDocs are restructured or retired, partner mappings need re-testing, and on-premises middleware that has quietly carried the business for fifteen years, invisible until the moment it failed, cannot keep up. The bill comes due. But the real choice is not how to fix the plumbing. It is whether the connectivity layer should remain a way of moving documents between two known parties, or become a queryable network that an autonomous agent can reason over. The answer the market is offering, namely modernise the EDI, move it to the cloud, add a bit of AI, is the corporate equivalent of printing out a route from the website. Technically more modern. Functionally the same piece of paper.

Give cloud EDI its due first. Moving an on-premises EDI estate into a managed cloud genuinely helps: better resilience, faster partner onboarding, centralised monitoring, and no more single-point-of-failure when your one EDI specialist retires. I am not arguing against any of it. I am arguing that none of it changes what the architecture is.

Cloud EDI is still EDI. It is still bilateral. It is still static. It produces a dead document, accurate at the moment of transmission, useless the moment anything changes, and it has nothing useful to offer an autonomous agent. Agents do not just process documents after the fact. They need to interrogate live, networked data, who has capacity now, at what price, under what constraints, and EDI has no way of exposing that.

This is the part that needs spelling out. Consider a worked example. A buyer agent at a Tier 1 manufacturer receives an order for three thousand units of a sub-assembly. To place that order well, the agent needs to know which suppliers across the network have capacity, at what price, with what lead time, tariff exposure, and carbon intensity. It needs to consider whether splitting the order across two suppliers reduces risk enough to justify the higher unit price, and whether either supplier is constrained by a Tier 2 component that has gone short this week. I have sat through several of those demos this year. Not one of them could have been accomplished over a one-to-one document pipe. Not one. The agent needs a network where it can ask a question and get an answer that reflects the current state of that network.

The architectural requirement is the same whatever you call it: networked, queryable, shared state across thousands of partners, with event-level visibility and the means to act. SAP Business Network is the specific instantiation of this within the S/4HANA ecosystem. The label matters less than the property. EDI was engineered in the 1970s to do the opposite of all of it.

Watch the market on this one. Read any vendor argument for “modernising your EDI to the cloud” carefully and you will find them describing a system that decouples mapping from the ERP, holds partner profiles centrally, and offers a real-time cockpit across thousands of partners. That is not modernised EDI. That is a network with a different name on the door. The market is quietly building the right thing while loudly selling the wrong word for it.

Two numbers tell you the shape of it. Interos, in its 6 Supply Chain Risks for 2026 report, finds that 90% of organisations remain blind to sub-tier disruptions for up to 48 hours. APQC's 2026 Supply Chain Priorities and Challenges: Cross-Industry Report finds that only one in three organisations have a strategy this year to increase visibility into lower-tier suppliers. Two-thirds are not even planning to do anything about it. One-to-one connections cannot reach those lower tiers. They cannot reach an agentic workflow either.

You already know I work in this space. The companies already on a network are building compound advantage in supplier visibility, agentic readiness, and network intelligence that a one-to-one pipe cannot replicate after the fact.

This should not be a Phase 2 conversation. It should not be kicked to 2028. It should not be left for the next team to discover. By then the programme team has moved on. The decision is business as usual. Nobody around the table remembers it was once a choice. A one-to-one pipe cannot reroute when a supplier fails, recompute when a tariff lands, or answer when an agent asks it a question. A network can do all three. The companies that defer will spend the next three years falling behind quietly, and in an agentic world three years is no longer a planning horizon. It is a generation. Kicking this to 2028 is not a deferral. It is a decision to fall behind, taken by people who will not be there to see it land.

The road atlas works. So does the printed page from the route planner. Neither of them is going to tell you about the lorry on the hard shoulder at the next junction, or that a faster route has just opened up that the rest of your competitors are already taking.