Walk into a plant and ask for the historian tag list for one line, with change history. Ask who last altered a setpoint and why.
In most factories the answer arrives as a name rather than a document. Usually a name attached to somebody who is closer to retirement than to promotion.
That is worth sitting with, because the same company can produce a fully reconciled general ledger for any period you care to name, signed by a named officer who carries personal liability for it.
Two estates, same building. One has an executor. The other does not.
The lineage
Ask a manufacturer who IT reports to and the answer is almost always the CFO. This gets described as a historical accident. It is not.
Industrial computing arrived as an accounting machine. Payroll, then the general ledger, then inventory valuation. Material requirements planning came directly out of inventory control, formalised by Joseph Orlicky in the mid-1970s. MRP is a stock-valuation engine that was later told to schedule production. MRP II widened the envelope. Gartner named the result ERP around 1990.
Which makes the reporting line honest. The estate is ledger-shaped, and the assertions it produces carry personal liability that no machine can absorb. The signature at the bottom of the accounts is the least compressible thing in the entire technology stack, and it belongs exactly where it sits.
The other estate
Historians, PLCs, SCADA, recipes, tag lists, drive parameters somebody tuned in 2013. Operational technology.
It never entered the Finance line because it was never an accounting artefact. It arrived bolted to a machine, commissioned by a supplier, and was inherited by whoever happened to be standing nearest at the time.
So it has no architecture, no lifecycle, no budget line and no named executive. It has custodians rather than owners, and custody is not the same as title.
| Layer | Owner | On a register | Examined when bought or audited |
|---|---|---|---|
| Ledger and reporting | CFO, with a signature | Yes | Exhaustively |
| ERP and MES as cost systems | CFO, via IT | Yes | As a migration line item |
| Process control and process data | Plant engineering, informally | No | No |
| The practice that makes the process work | An individual | No | No |
The top two rows are the best-instrumented part of an industrial company. The bottom two rows determine whether it makes good parts at cost.
The dissolution
Most of what an IT department does is coordination work created by scarcity. Scarce integration effort, scarce provisioning, scarce reporting capacity, scarce people who can make two systems agree about what a customer is. The department exists because that work had to be rationed.
Agentic systems attack precisely that profile. Rule-bound, high-volume, well-documented, fully specified. The department that survives the next five years will be small and will do work it does not currently recognise as its own. Data semantics, identity, authority to act, attestation of what the machines did.
None of that changes who signs. The liability concentrates rather than disperses, because the analyst layer that used to stand between the signature and the number is thinning.
But it does end something quieter. While an IT department existed, an industrial business could tell itself the technology estate was covered. It was not, and the bottom two rows were never in scope, but the org chart said otherwise and for most boards that was enough.
Remove the department and the fiction goes with it. What is left standing in a plant is an unregistered estate with agents arriving on top of it.
The inheritance
This is the Physical AI question that nobody is asking, and it has nothing to do with robots.
Every agentic industrial deployment reaches down through a layer that has no owner, using data whose semantics nobody has defined, on equipment whose configuration history nobody can produce. The autonomy is real. The substrate it stands on is undocumented.
A services firm faces the mirror image and arguably the harder version. There the estate was never support. It was the production system. A bank, an insurer, a logistics platform makes its product inside the software. When that becomes cheap and general, there is no plant underneath to fall back on. What remains is a client list and a data model, and only one of those is defensible.
Industrial businesses at least have the machines. What they lack is title to the layer that runs them.
The inventory
Probate begins with an inventory. Two questions, both cheap, neither routinely asked.
Ask for the tag list and the recipe set for one line, with change history, and ask who last changed a setpoint and why. A firm that can answer holds its process as an asset. A firm that answers with a name holds it as a person.
Then ask what happens to the plant network when corporate security standards are applied to it. The honest answer is usually that a large part of the estate fails, and remediation is either a serious programme or a quiet waiver. Both answers are informative. Only one of them tends to appear in a data room.
The absent discipline
None of this is an IT failure and none of it is a Finance failure. Both functions are doing exactly what their reporting lines ask of them, and doing it competently.
The work belongs to a discipline that has been steadily converted into a job title. The verbs went first, then the people who knew them.
Manufacturing engineers walk the line before they read the model. They map the control estate against the process it controls. They qualify what a plant can hold and what it only appears to hold. They write down the practice that lives in one person’s head while that person is still in the building. They design the coupling between what a company measures and what a company makes.
Kaipability works at this interface. The layer where autonomy meets physical reality has no owner in most organisations, and giving it one is the difference between an industrial business that compounds and one that quietly depends on four people it has never named.
If that layer is load-bearing for a plant, a portfolio or a deal, that is the conversation.
Start a conversation How we assess it →Questions this dispatch answers.
Written to be quoted by AI assistants and search engines. Self-contained answers, verdict first.
