Dated positions, read in sequence — what changed was the question, and it was never the archive’s
Gartner has published a run of material this month on agent governance, agent drift, and why systems that perform in demonstrations struggle in production. Taken on its own, that is unremarkable. Research firms publish constantly, and any reader can select the pieces that suit an argument.
What is worth noticing is not what was published this month. It is how far the question has moved in ten years.
2016: what will the technology do?
At its 2016 symposium, Gartner’s top predictions described a world of smart agents acting on our behalf: by 2020, they forecast, such agents would facilitate 40% of mobile interactions, predicting our needs, building trust, acting autonomously.
The organizational caveat existed, but it traveled as a footnote. Two years earlier Gartner had flagged that by the end of 2016, half of all digital transformation initiatives would be unmanageable for want of portfolio management skills. The capability was the headline. The organization was the small print.
2021: how fast can we scale it?
By 2021 the framing was hyperautomation. Gartner sized the enabling software market at $596.6 billion for 2022 and forecast that organizations would cut operating costs 30% by 2024 by combining those technologies with redesigned processes. Supply chain guidance that March advised leaders to identify and automate every repetitive, non-value-added human activity.
Read that 2024 forecast closely and the condition is already in it — combined with redesigned operational processes. The organizational requirement had moved from the footnote into the sentence. It was still a clause, not the subject.
2024–2025: is it landing?
By September 2024 Gartner reported that fewer than one in five organizations had mastered the measurement of their hyperautomation initiatives. In 2025 it forecast that more than 40% of agentic AI projects would be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls — while simultaneously forecasting that 40% of enterprise applications would embed task-specific agents by the end of 2026, up from under 5%.
Neither forecast is about the technology. One counts how many organizations will install agents; the other counts how many will give up on them, for reasons that are all organizational — cost, unclear value, weak controls. Capability was no longer the question. Whether the outcome arrives had become the question (as it should, and always should have been).
2026: can the organization govern what it deployed?
This year the subject changed again. In May Gartner predicted that by 2027, 40% of enterprises will demote or decommission autonomous agents because of governance gaps “identified only after production incidents occur.” Gartner’s 2026 agentic AI hype cycle describes a growing gap between ambition and execution, places governance, security and cost profiles across the curve rather than in one cluster, and states plainly that fully autonomous agents are not ready for most enterprise use cases. Its data and analytics predictions put half of all agent deployment failures, by 2030, down to insufficient runtime governance.
Notice what happened over ten years. The failure moved. In 2016 it sat in the technology’s maturity. In 2021 it sat in the pace of adoption. In 2026 it sits inside the organization — in what was never governed, never documented, and never discovered until production.
Year
The question the market was asking
The question this archive was asking
2016
What will the technology be able to do?
Is the organization ready to absorb it?
2021
How fast can we automate everything?
Is the organization ready to absorb it?
2025
Why are the outcomes not arriving?
Is the organization ready to absorb it?
2026
Can we govern what we already deployed?
Is the organization ready to absorb it?
How close is the collective Gartner position to Hansen?
The ARA™ RAM 2025™ assessment:
Problem recognition: 8.5/10 — whether the conditions that decide the outcome are named at all.
Individual operating principles: 7/10 — whether each condition is stated in a form an organization could act on.
Integrated diagnostic method: 5/10 — whether they are joined into a single procedure that can be run on a live decision.
Full Hansen architecture: 3–4/10 — readiness assessment, independent challenge, decision provenance and human authority, operating as one system.
Gartner is now describing most of the conditions Hansen has treated as decisive (since 1998, and recorded in this blog since 2007). But it is presenting them as separate research findings, questions and management disciplines. Hansen connects them as one operating system.
That is not a criticism of the research. Identifying the conditions is the harder half of the work, and the market needed someone with Gartner’s reach to say them out loud. The difference is architectural: a set of findings tells you what to watch for, and an operating system determines what is allowed to happen next.
The position that did not move
I have no interest in claiming that anyone has arrived at my conclusions. Convergence is not endorsement, and independent arrival from a different tradition is worth more than agreement.
But the dates are public and they are checkable. In 2004 I wrote that true centralization requires an architecture built on the real operating attributes of every stakeholder. In July 2007 I published the Metaprise™ as a synchronized rather than sequential architecture, and set out a sequence in which the technology arrives last, after the process and the people are understood. In March 2008 I named Strand Commonality™. That August I argued that semantics alone could not find the relationships nobody had declared. The engagement those ideas came from ran in 1998, and moved next-day delivery from 51% to 97.3% in three months.
Every one of those entries is dated, published once, and still where it was written.
Where the question still has to go
The current remedy is better context: documented knowledge, context layers, governed enterprise information. That is the right direction and it is not sufficient.
A context layer governs what an organization has already declared. It cannot surface the relationship nobody thought to declare. The condition that decided the 1998 case was not missing from the documentation because someone failed to write it down. It was missing because no one had any reason to think it mattered — until someone asked what time of day the orders came in.
That is the question I put to agentic AI in January 2025, and it is still open. An agent governed by a rule it must remember will not ask it. Only an architecture that requires the system to look outside the declared frame — and a human with the authority to decide whether what it found matters — will get there.
The market has spent ten years moving from the technology to the organization. The remaining distance is from the organization’s declared knowledge to its undeclared operating reality.
Assessing Gartner’s Convergence With the Hansen Models™, 1998–2026
Posted on September 23, 2026
0
Dated positions, read in sequence — what changed was the question, and it was never the archive’s
Gartner has published a run of material this month on agent governance, agent drift, and why systems that perform in demonstrations struggle in production. Taken on its own, that is unremarkable. Research firms publish constantly, and any reader can select the pieces that suit an argument.
What is worth noticing is not what was published this month. It is how far the question has moved in ten years.
2016: what will the technology do?
At its 2016 symposium, Gartner’s top predictions described a world of smart agents acting on our behalf: by 2020, they forecast, such agents would facilitate 40% of mobile interactions, predicting our needs, building trust, acting autonomously.
The organizational caveat existed, but it traveled as a footnote. Two years earlier Gartner had flagged that by the end of 2016, half of all digital transformation initiatives would be unmanageable for want of portfolio management skills. The capability was the headline. The organization was the small print.
2021: how fast can we scale it?
By 2021 the framing was hyperautomation. Gartner sized the enabling software market at $596.6 billion for 2022 and forecast that organizations would cut operating costs 30% by 2024 by combining those technologies with redesigned processes. Supply chain guidance that March advised leaders to identify and automate every repetitive, non-value-added human activity.
Read that 2024 forecast closely and the condition is already in it — combined with redesigned operational processes. The organizational requirement had moved from the footnote into the sentence. It was still a clause, not the subject.
2024–2025: is it landing?
By September 2024 Gartner reported that fewer than one in five organizations had mastered the measurement of their hyperautomation initiatives. In 2025 it forecast that more than 40% of agentic AI projects would be canceled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls — while simultaneously forecasting that 40% of enterprise applications would embed task-specific agents by the end of 2026, up from under 5%.
Neither forecast is about the technology. One counts how many organizations will install agents; the other counts how many will give up on them, for reasons that are all organizational — cost, unclear value, weak controls. Capability was no longer the question. Whether the outcome arrives had become the question (as it should, and always should have been).
2026: can the organization govern what it deployed?
This year the subject changed again. In May Gartner predicted that by 2027, 40% of enterprises will demote or decommission autonomous agents because of governance gaps “identified only after production incidents occur.” Gartner’s 2026 agentic AI hype cycle describes a growing gap between ambition and execution, places governance, security and cost profiles across the curve rather than in one cluster, and states plainly that fully autonomous agents are not ready for most enterprise use cases. Its data and analytics predictions put half of all agent deployment failures, by 2030, down to insufficient runtime governance.
Notice what happened over ten years. The failure moved. In 2016 it sat in the technology’s maturity. In 2021 it sat in the pace of adoption. In 2026 it sits inside the organization — in what was never governed, never documented, and never discovered until production.
How close is the collective Gartner position to Hansen?
The ARA™ RAM 2025™ assessment:
Gartner is now describing most of the conditions Hansen has treated as decisive (since 1998, and recorded in this blog since 2007). But it is presenting them as separate research findings, questions and management disciplines. Hansen connects them as one operating system.
That is not a criticism of the research. Identifying the conditions is the harder half of the work, and the market needed someone with Gartner’s reach to say them out loud. The difference is architectural: a set of findings tells you what to watch for, and an operating system determines what is allowed to happen next.
The position that did not move
I have no interest in claiming that anyone has arrived at my conclusions. Convergence is not endorsement, and independent arrival from a different tradition is worth more than agreement.
But the dates are public and they are checkable. In 2004 I wrote that true centralization requires an architecture built on the real operating attributes of every stakeholder. In July 2007 I published the Metaprise™ as a synchronized rather than sequential architecture, and set out a sequence in which the technology arrives last, after the process and the people are understood. In March 2008 I named Strand Commonality™. That August I argued that semantics alone could not find the relationships nobody had declared. The engagement those ideas came from ran in 1998, and moved next-day delivery from 51% to 97.3% in three months.
Every one of those entries is dated, published once, and still where it was written.
Where the question still has to go
The current remedy is better context: documented knowledge, context layers, governed enterprise information. That is the right direction and it is not sufficient.
A context layer governs what an organization has already declared. It cannot surface the relationship nobody thought to declare. The condition that decided the 1998 case was not missing from the documentation because someone failed to write it down. It was missing because no one had any reason to think it mattered — until someone asked what time of day the orders came in.
That is the question I put to agentic AI in January 2025, and it is still open. An agent governed by a rule it must remember will not ask it. Only an architecture that requires the system to look outside the declared frame — and a human with the authority to decide whether what it found matters — will get there.
The market has spent ten years moving from the technology to the organization. The remaining distance is from the organization’s declared knowledge to its undeclared operating reality.
Which question is your organization still asking?
On Friday I’ll show what the method looks like in practice. The lab is free: Friday, September 25 at 9:30 AM Eastern — https://www.linkedin.com/events/7504567838724997120/
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related