Jon W. Hansen, FCIPS · Procurement Insights | Hansen Models™ · 16 October 2026
On 13 October, McKinsey told CEOs that the answer to stalled AI value is an AI operating system. Its article, The AI operating system: A new way for CEOs to run the enterprise, describes a continuously operating management layer that holds enterprise context, reasons across functions, connects the front line to the executive suite, and learns from outcomes. It frames the problem with a striking gap from its 2026 State of AI survey: 40 percent of respondents at large companies report scaling AI agents, yet only 6 percent say AI generates at least 5 percent of EBIT.
McKinsey deserves credit for putting the idea in front of the people who decide whether it gets built. That matters, and it moves the conversation where it needs to go.
It is not, however, a new way. In 1998, an SR&ED-funded system built on Strand Commonality™ and an agent-based architecture, later named Metaprise™, was already running in a live operation, its self-learning algorithms feeding every outcome back into the next decision: the firm fences of a nascent AI framework. What is new in 2026 is the models and the scale. The operating logic is twenty-eight years old, and its record is public, dated and still online.
The record, in order
In 1998, it was built and running. With funding from the Government of Canada’s Scientific Research and Experimental Development program, I built an agent-based system for a defence maintenance operation supporting the Department of National Defence. Next-day parts delivery stood at 51 percent against a 90 percent contractual requirement. The system linked every party that touched the outcome, from service technicians and buyers to suppliers, couriers, customs and finance, and its self-learning algorithms fed each outcome back into the next decision. Delivery reached 97.3 percent within three months and held for seven years. It was not a pilot or a demonstration: the system ran under a live, multimillion-dollar contract, with the clock running.
In 2001, its architecture was published. The core of that architecture was documented in international patent application WO 01/65428 A2, Interactive Parts Ordering Management Method and System, published on 7 September 2001.
Figure 1 of international patent application WO 01/65428 A2, published 7 September 2001: the system architecture, with administrators, sourcers and customer service representatives designed in as part of the system, and the ERP as one component rather than the center.
In 2004, the operating principle was written down. My paper Acres of Diamonds stated that “a true centralization of procurement objectives requires a decentralized architecture that is based on the real-world operating attributes of all transactional stakeholders,” and called that the cornerstone of agent-based modeling. A 2005 paper described the need for a continuous record of an operation in motion, a camcorder rather than a snapshot, capturing what walks into the frame that nobody put there.
In 2007, it went on the public record. The 2004 principle appeared on Procurement Insights in December 2007.
In 2008, it met the digital nervous system. In April 2008, a reader asked whether organizations had the “digital nervous system” that Bill Gates and Collins Hemingway described in their 1999 book, Business @ the Speed of Thought. It was my first encounter with the book. My answer set out why agent-based models that begin with real operating attributes could deliver what the vendors’ architectures could not. Three days later, the 2005 camcorder passage was published.
In May 2025, it was named. A Procurement Insights post stated that generative AI requires a central AI operating system, and that the 1998 framework provided the foundational logic for one, adding that while such a system involves technology, it is not a technology-first undertaking.
In September 2026, the readiness question was asked directly. When McKinsey argued that continuous, real-time decision-making was finally possible, I agreed, and asked whether organizations were any more ready for it than they were in 1998.
The name changed with the era; the idea did not. From 2 June 2007, this archive called the architecture Metaprise™, and a search of the archive for that name returns 368 items. In 2024, as the market adopted the phrase “AI operating system,” the archive did too, and a search for that phrase returns 34. Across both names, the same operating problem recurs: how to connect dynamic operating relationships, learn from outcomes, and prevent technology from making a mistaken understanding of the operation faster and more authoritative.
These counts cover the Procurement Insights blog archive only. They do not include the podcast archive, the white paper and presentation archive, or the hard-copy records, which include the 2004 and 2005 papers referenced above.
This is not a claim to every idea now associated with an AI operating system. It is a documented operating record: a working framework of what is now called an enterprise AI operating system was applied in a live environment in 1998, published in an international patent application in 2001, written about in papers dated 2004 and 2005, and on the public record through Procurement Insights from 2007 onward.
The distinction that the record carries
There is one difference between the operating record and the current framing, and it is the one that matters most.
An AI operating system is built from enterprise context: the data, rules, policies, workflows and institutional knowledge an organization has accumulated. Incumbents, the argument goes, have an advantage in decades of that context.
But accumulated context is the organization’s remembered explanation of itself, and that explanation can be wrong. In 1998, nothing in the formal process maps, transaction records or individual performance measures revealed that service technicians were holding part orders until late afternoon to protect a call-response target. That behavior set off a chain of effects through sourcing, cross-border delivery, courier selection, cost and finance. Codifying the established rules and records would have made the wrong explanation faster, more consistent and more authoritative.
That is the problem Phase 0™ exists to resolve: testing whether the context corresponds to operating reality, and whether contradictory outcomes are allowed to change it, before the operating system compounds it.
December 2025 (Phase 0™) – https://procureinsights.com/2025/12/22/the-case-for-phase-0-industry-voices-on-readiness-first-transformation/
Where the line sits
The record above is public. Every post is dated and still online, the 1998 results are documented, and the patent application is a matter of public record. Anyone can check the chronology.
The operating method is a different matter. How the trace is done, how the strands are identified and how the system tests its own explanation remain proprietary, and they are available through engagements, not through posts.
The evidence is public. The operating method is proprietary.
The market is assembling the modern AI operating system in 2026. It is not starting from an empty historical record. The question the current framing has not yet answered is the one that came first in 1998: before you encode what your organization believes about itself, have you confirmed that it reflects how the operation actually works in the real world, from the enterprise frontlines to the boardroom?
The record:
The sources:
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Related
Before the AI Operating System Had a Name, It Had an Operating Record
Posted on October 16, 2026
0
Jon W. Hansen, FCIPS · Procurement Insights | Hansen Models™ · 16 October 2026
On 13 October, McKinsey told CEOs that the answer to stalled AI value is an AI operating system. Its article, The AI operating system: A new way for CEOs to run the enterprise, describes a continuously operating management layer that holds enterprise context, reasons across functions, connects the front line to the executive suite, and learns from outcomes. It frames the problem with a striking gap from its 2026 State of AI survey: 40 percent of respondents at large companies report scaling AI agents, yet only 6 percent say AI generates at least 5 percent of EBIT.
McKinsey deserves credit for putting the idea in front of the people who decide whether it gets built. That matters, and it moves the conversation where it needs to go.
It is not, however, a new way. In 1998, an SR&ED-funded system built on Strand Commonality™ and an agent-based architecture, later named Metaprise™, was already running in a live operation, its self-learning algorithms feeding every outcome back into the next decision: the firm fences of a nascent AI framework. What is new in 2026 is the models and the scale. The operating logic is twenty-eight years old, and its record is public, dated and still online.
The record, in order
In 1998, it was built and running. With funding from the Government of Canada’s Scientific Research and Experimental Development program, I built an agent-based system for a defence maintenance operation supporting the Department of National Defence. Next-day parts delivery stood at 51 percent against a 90 percent contractual requirement. The system linked every party that touched the outcome, from service technicians and buyers to suppliers, couriers, customs and finance, and its self-learning algorithms fed each outcome back into the next decision. Delivery reached 97.3 percent within three months and held for seven years. It was not a pilot or a demonstration: the system ran under a live, multimillion-dollar contract, with the clock running.
In 2001, its architecture was published. The core of that architecture was documented in international patent application WO 01/65428 A2, Interactive Parts Ordering Management Method and System, published on 7 September 2001.
Figure 1 of international patent application WO 01/65428 A2, published 7 September 2001: the system architecture, with administrators, sourcers and customer service representatives designed in as part of the system, and the ERP as one component rather than the center.
In 2004, the operating principle was written down. My paper Acres of Diamonds stated that “a true centralization of procurement objectives requires a decentralized architecture that is based on the real-world operating attributes of all transactional stakeholders,” and called that the cornerstone of agent-based modeling. A 2005 paper described the need for a continuous record of an operation in motion, a camcorder rather than a snapshot, capturing what walks into the frame that nobody put there.
In 2007, it went on the public record. The 2004 principle appeared on Procurement Insights in December 2007.
In 2008, it met the digital nervous system. In April 2008, a reader asked whether organizations had the “digital nervous system” that Bill Gates and Collins Hemingway described in their 1999 book, Business @ the Speed of Thought. It was my first encounter with the book. My answer set out why agent-based models that begin with real operating attributes could deliver what the vendors’ architectures could not. Three days later, the 2005 camcorder passage was published.
In May 2025, it was named. A Procurement Insights post stated that generative AI requires a central AI operating system, and that the 1998 framework provided the foundational logic for one, adding that while such a system involves technology, it is not a technology-first undertaking.
In September 2026, the readiness question was asked directly. When McKinsey argued that continuous, real-time decision-making was finally possible, I agreed, and asked whether organizations were any more ready for it than they were in 1998.
The name changed with the era; the idea did not. From 2 June 2007, this archive called the architecture Metaprise™, and a search of the archive for that name returns 368 items. In 2024, as the market adopted the phrase “AI operating system,” the archive did too, and a search for that phrase returns 34. Across both names, the same operating problem recurs: how to connect dynamic operating relationships, learn from outcomes, and prevent technology from making a mistaken understanding of the operation faster and more authoritative.
These counts cover the Procurement Insights blog archive only. They do not include the podcast archive, the white paper and presentation archive, or the hard-copy records, which include the 2004 and 2005 papers referenced above.
This is not a claim to every idea now associated with an AI operating system. It is a documented operating record: a working framework of what is now called an enterprise AI operating system was applied in a live environment in 1998, published in an international patent application in 2001, written about in papers dated 2004 and 2005, and on the public record through Procurement Insights from 2007 onward.
The distinction that the record carries
There is one difference between the operating record and the current framing, and it is the one that matters most.
An AI operating system is built from enterprise context: the data, rules, policies, workflows and institutional knowledge an organization has accumulated. Incumbents, the argument goes, have an advantage in decades of that context.
But accumulated context is the organization’s remembered explanation of itself, and that explanation can be wrong. In 1998, nothing in the formal process maps, transaction records or individual performance measures revealed that service technicians were holding part orders until late afternoon to protect a call-response target. That behavior set off a chain of effects through sourcing, cross-border delivery, courier selection, cost and finance. Codifying the established rules and records would have made the wrong explanation faster, more consistent and more authoritative.
That is the problem Phase 0™ exists to resolve: testing whether the context corresponds to operating reality, and whether contradictory outcomes are allowed to change it, before the operating system compounds it.
December 2025 (Phase 0™) – https://procureinsights.com/2025/12/22/the-case-for-phase-0-industry-voices-on-readiness-first-transformation/
Where the line sits
The record above is public. Every post is dated and still online, the 1998 results are documented, and the patent application is a matter of public record. Anyone can check the chronology.
The operating method is a different matter. How the trace is done, how the strands are identified and how the system tests its own explanation remain proprietary, and they are available through engagements, not through posts.
The evidence is public. The operating method is proprietary.
The market is assembling the modern AI operating system in 2026. It is not starting from an empty historical record. The question the current framing has not yet answered is the one that came first in 1998: before you encode what your organization believes about itself, have you confirmed that it reflects how the operation actually works in the real world, from the enterprise frontlines to the boardroom?
The record:
The sources:
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related