Agentic execution is not the new problem. The enduring problem is how the system discovers what matters, assigns decision rights correctly, and lets real-world outcomes correct the next action.
Jon W. Hansen, FCIPS | Procurement Insights | August 2026
THE SHORT VERSION FOR BUSY EXECUTIVES
For the past week I have been writing about where AI belongs in a procurement decision, when to examine a process that appears to be working, and why the sequence that puts operating reality downstream of the architecture has survived four technology eras.
All of it rested on a case I ran in 1998 and have described consistently ever since. Documented, but described by me.
This week a different kind of document surfaced. Not my account of the architecture — the architecture, examined and published by an international patent authority.
WO 01/65428 A2, Interactive Parts Ordering Management Method and System. Published 7 September 2001. Priority filings 2 March 2000 in Canada and 28 March 2000 in the United States.
It describes software agents that are autonomous, pro-active and reactive. It claims intelligent time zone polling across inventory locations. It claims a system that monitors whether a shipment reported back and takes progressive action when it did not.
WIPO published it. Anyone has been able to read it ever since.
That is an agentic procurement loop, specified in the language of agents, filed twenty-six years ago.
The part that should interest you is not that it exists. It is that it was published. The specification has been sitting in the public record, readable by anyone, for a quarter of a century.
Twenty-six years later, the implementation failure rate has not moved.
A DEEPER DIVE
What the document actually says
I am quoting the published specification rather than characterizing it.
On the agent, at page 6:
An interactive service agent 14 is provided in the system 10 to permit and control real-time communications between the user 30 and a customer service representative (CSR) 45, the agent being a segment of computer software which can be autonomous and/or mobile and may be implemented as a component or object. (Agents are able to interact with their environment and to act both pro-actively and reactively).
Read the parenthesis again. That is a definition of agency, written in 2000, in a filing that had to be precise enough for an examiner.
On the polling, from the claims:
…means for determining from said identified inventory locations a target inventory location from which said goods are deliverable to said destination address within said time delivery requirement… means for intelligent time zone polling of said identified inventory locations.
Figure 1, WO 01/65428 A2 (published 7 September 2001). Reproduced from the published international application. Note the three separate CPUs — A, B and C — and Web Tracker at the right-hand edge.
And the specification works the example. If the local warehouse does not have the part, the system moves to the nearest warehouse by distance in the same time zone, recalculates the delivery capability, and either confirms the service-level requirement can be met or returns the achievable alternative.
And then the claim that matters most:
…means for receiving shipment information from said target inventory location, means for monitoring whether shipment information for said order has been received and means for taking progressive action with respect to said order if such shipment information is not received within a predetermined time period.
Monitoring against an expected result. Acting on the gap. Not a dashboard. Not a report. An escalation triggered by an absence.
That is what most organizations are trying to buy right now.
What it was competing against
The background section of the same document describes the state of the art it was filed against. Parts brokers, it says, had applied manual systems, using pager and fax communications, with order-by-order calculations and decisions, for checking warehouse inventory, selecting a warehouse and initiating shipping. Personnel-dependent. Prone to error.
That is the baseline. Pagers and fax machines.
Against that, a specification describing autonomous agents and automated escalation reads as speculative — which is roughly how it was received. And you can see the reception in the drafting itself. The document stops to explain what a software agent is, and defines component as a set of computer-readable instructions. Nobody defines a term the reader already knows.
The vocabulary was not established. The thing it described was working.
Why publication is the interesting part
The patent system runs on a bargain: an applicant discloses how the thing works, in enough detail that someone skilled in the field could build it, and in exchange may obtain a period of protection. Disclosure is not a side effect of the process — it is the price of admission.
Which means the specification has been in the public record since September 2001. Not summarized. Not described. The working detail, published by an international authority, free to read.
I make no claim here about its current legal status. I have not checked, and it does not matter to the point. The disclosure was always public.
And across the twenty-five years since, the industry has spent very large sums pursuing:
- systems that determine the best fulfilment source against a service-level requirement
- coordination across distributed locations with different local conditions
- automated initiation of downstream execution
- and monitoring that escalates when an expected result does not arrive
All four are described in a public document that has cost nothing to read since 2001.
I am not claiming anyone should have found it. Nobody trawls twenty-five-year-old patent publications for architecture, and I would not have expected them to. The point is not neglect — it is that the knowledge was never the constraint.
Which brings me back to a question I was asked this week
A researcher studying multi-agent AI deployments in enterprise supply chains published findings suggesting multi-agent architectures show a stronger association with recorded success than single-agent alternatives. He was careful about it — he disclosed the confound himself, that multi-agent systems tend to get chosen for problems where coordination is the point, and those problems may be structurally easier to demonstrate a win on.
I put a table to him: ten steps of the 1998 loop, what the architecture did, and where I believe modern AI contributes.
Here is the sharper version of the question, now that the filing has surfaced.
The agents in that system were the transactional stakeholders — suppliers, couriers, customs brokers, service, finance — coordinated by deterministic computation and self-learning algorithms. Next-day delivery went from fifty-one percent to ninety-seven point three in three months. Within eighteen months the collective buying groups went from twenty-three FTE equivalents to three.
No large language model. No generative capability. None of it existed.
So the question is not whether AI helps. It does, and I have said where: reading the world outside the transaction feed, parsing unstructured service records, reconciling part numbering continuously rather than as a project. Those are real gains and I would take all of them.
The question is why anyone believes AI is the enabling condition.
The architecture did not require today’s AI to succeed. Modern AI can make several of its functions cheaper, wider, faster and continuous — but that is a different claim from saying AI created the underlying operating architecture.
What this does to the last four posts
On the operating system: I argued that the architecture was right and nobody had said where its intelligence layer comes from. One answer was filed, examined, published and released into the commons. The question was answerable. It was not asked.
On the Human/Tech Stack™: I argued that placement matters more than capability. The claims are placement, written in legal language — what determines the target, what initiates shipment, what monitors and escalates, and what the human still decides.
On disturbing success: the clean number the industry currently reports is we are building something genuinely new. The filing says otherwise, and nothing in anyone’s process was going to surface that.
And on the 1983 decision hierarchy: the sequence has not changed. Now there is a dated public record at both ends — pager-and-fax as the documented state of the art in 2000, and an implementation failure rate that has not moved in 2026.
The claim I am actually making
Not that I predicted anything. Prediction assumes the thing did not exist yet.
What decides procurement outcomes is not a single link between two things. It is multiple points of simultaneous collective coupling — a set of conditions in different parts of the operation bearing on the same decision at once. That was operating in 1998, is operating now, and was operating through every intervening era. What changed is whether anyone traced it. The filing is not a forecast that came true — it is a second, independent record of the same constraint, made by someone who had gone and looked.
Availability was never the problem. Neither was capability.
The problem is that nobody establishes what the operation is actually doing before they decide what to build on top of it. That was true when the alternative was a fax machine, and it is true now that the alternative is an Agentic AI agent.
Jon W. Hansen, FCIPS — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Related
The Agents Were Filed in 2000. The Architecture Was Already Working.
Posted on August 28, 2026
0
Agentic execution is not the new problem. The enduring problem is how the system discovers what matters, assigns decision rights correctly, and lets real-world outcomes correct the next action.
Jon W. Hansen, FCIPS | Procurement Insights | August 2026
THE SHORT VERSION FOR BUSY EXECUTIVES
For the past week I have been writing about where AI belongs in a procurement decision, when to examine a process that appears to be working, and why the sequence that puts operating reality downstream of the architecture has survived four technology eras.
All of it rested on a case I ran in 1998 and have described consistently ever since. Documented, but described by me.
This week a different kind of document surfaced. Not my account of the architecture — the architecture, examined and published by an international patent authority.
WO 01/65428 A2, Interactive Parts Ordering Management Method and System. Published 7 September 2001. Priority filings 2 March 2000 in Canada and 28 March 2000 in the United States.
It describes software agents that are autonomous, pro-active and reactive. It claims intelligent time zone polling across inventory locations. It claims a system that monitors whether a shipment reported back and takes progressive action when it did not.
WIPO published it. Anyone has been able to read it ever since.
That is an agentic procurement loop, specified in the language of agents, filed twenty-six years ago.
The part that should interest you is not that it exists. It is that it was published. The specification has been sitting in the public record, readable by anyone, for a quarter of a century.
Twenty-six years later, the implementation failure rate has not moved.
A DEEPER DIVE
What the document actually says
I am quoting the published specification rather than characterizing it.
On the agent, at page 6:
Read the parenthesis again. That is a definition of agency, written in 2000, in a filing that had to be precise enough for an examiner.
On the polling, from the claims:
Figure 1, WO 01/65428 A2 (published 7 September 2001). Reproduced from the published international application. Note the three separate CPUs — A, B and C — and Web Tracker at the right-hand edge.
And the specification works the example. If the local warehouse does not have the part, the system moves to the nearest warehouse by distance in the same time zone, recalculates the delivery capability, and either confirms the service-level requirement can be met or returns the achievable alternative.
And then the claim that matters most:
Monitoring against an expected result. Acting on the gap. Not a dashboard. Not a report. An escalation triggered by an absence.
That is what most organizations are trying to buy right now.
What it was competing against
The background section of the same document describes the state of the art it was filed against. Parts brokers, it says, had applied manual systems, using pager and fax communications, with order-by-order calculations and decisions, for checking warehouse inventory, selecting a warehouse and initiating shipping. Personnel-dependent. Prone to error.
That is the baseline. Pagers and fax machines.
Against that, a specification describing autonomous agents and automated escalation reads as speculative — which is roughly how it was received. And you can see the reception in the drafting itself. The document stops to explain what a software agent is, and defines component as a set of computer-readable instructions. Nobody defines a term the reader already knows.
The vocabulary was not established. The thing it described was working.
Why publication is the interesting part
The patent system runs on a bargain: an applicant discloses how the thing works, in enough detail that someone skilled in the field could build it, and in exchange may obtain a period of protection. Disclosure is not a side effect of the process — it is the price of admission.
Which means the specification has been in the public record since September 2001. Not summarized. Not described. The working detail, published by an international authority, free to read.
I make no claim here about its current legal status. I have not checked, and it does not matter to the point. The disclosure was always public.
And across the twenty-five years since, the industry has spent very large sums pursuing:
All four are described in a public document that has cost nothing to read since 2001.
I am not claiming anyone should have found it. Nobody trawls twenty-five-year-old patent publications for architecture, and I would not have expected them to. The point is not neglect — it is that the knowledge was never the constraint.
Which brings me back to a question I was asked this week
A researcher studying multi-agent AI deployments in enterprise supply chains published findings suggesting multi-agent architectures show a stronger association with recorded success than single-agent alternatives. He was careful about it — he disclosed the confound himself, that multi-agent systems tend to get chosen for problems where coordination is the point, and those problems may be structurally easier to demonstrate a win on.
I put a table to him: ten steps of the 1998 loop, what the architecture did, and where I believe modern AI contributes.
Here is the sharper version of the question, now that the filing has surfaced.
The agents in that system were the transactional stakeholders — suppliers, couriers, customs brokers, service, finance — coordinated by deterministic computation and self-learning algorithms. Next-day delivery went from fifty-one percent to ninety-seven point three in three months. Within eighteen months the collective buying groups went from twenty-three FTE equivalents to three.
No large language model. No generative capability. None of it existed.
So the question is not whether AI helps. It does, and I have said where: reading the world outside the transaction feed, parsing unstructured service records, reconciling part numbering continuously rather than as a project. Those are real gains and I would take all of them.
The question is why anyone believes AI is the enabling condition.
The architecture did not require today’s AI to succeed. Modern AI can make several of its functions cheaper, wider, faster and continuous — but that is a different claim from saying AI created the underlying operating architecture.
What this does to the last four posts
On the operating system: I argued that the architecture was right and nobody had said where its intelligence layer comes from. One answer was filed, examined, published and released into the commons. The question was answerable. It was not asked.
On the Human/Tech Stack™: I argued that placement matters more than capability. The claims are placement, written in legal language — what determines the target, what initiates shipment, what monitors and escalates, and what the human still decides.
On disturbing success: the clean number the industry currently reports is we are building something genuinely new. The filing says otherwise, and nothing in anyone’s process was going to surface that.
And on the 1983 decision hierarchy: the sequence has not changed. Now there is a dated public record at both ends — pager-and-fax as the documented state of the art in 2000, and an implementation failure rate that has not moved in 2026.
The claim I am actually making
Not that I predicted anything. Prediction assumes the thing did not exist yet.
What decides procurement outcomes is not a single link between two things. It is multiple points of simultaneous collective coupling — a set of conditions in different parts of the operation bearing on the same decision at once. That was operating in 1998, is operating now, and was operating through every intervening era. What changed is whether anyone traced it. The filing is not a forecast that came true — it is a second, independent record of the same constraint, made by someone who had gone and looked.
Availability was never the problem. Neither was capability.
The problem is that nobody establishes what the operation is actually doing before they decide what to build on top of it. That was true when the alternative was a fax machine, and it is true now that the alternative is an Agentic AI agent.
Jon W. Hansen, FCIPS — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related