The Human/Tech Stack™
Jon Hansen | Procurement Insights | August 2026
THE SHORT VERSION FOR BUSY EXECUTIVES
Almost every published account of AI in procurement shows where AI goes. Very few show where it does not go, and none that I have seen explain why.
We know where AI belongs in a procurement decision and where it does not, and we can show you which is which before you commit to anything.
That comes from an architecture built in 1998 that ran for seven years — self-learning algorithms on the hardware of the period. The Human/Tech Stack™ shows what twenty-eight years of capability since has changed, and what it has not.
Ten elements of a working procurement operation. For each one: who executes it, what it delivers, and what actually moved.
Nine of the ten existed in 1998. Four are unchanged. One is genuinely new. And neither of the two elements that produced the result — the ranking and the weighting — is one that AI touches.
That is not an argument against AI. Five of the ten are the same work made cheaper, wider or continuous, and those gains are real. It is an argument about placement: the elements that decide an outcome are not the elements where capability improved.
The Human/Tech Stack™ — Hansen Models™ | Procurement Insights
How to read it
Four columns. The element. Who executes it — and every executor is named as an agent class, because a procurement operation has always run on agents: internal humans, external humans, deterministic computation, and now an AI agent. What that element delivers. And what twenty-eight years of capability improvement did to it.
The left bar and the right column measure different things, and it is worth knowing which is which. The bar says whether the core architecture is unchanged — four solid, six dashed. The right-hand column says what AI did to the work: unchanged, assists, reduces cost, expands, or new.
They are not the same axis. Six of the dashed rows existed in 1998; what changed is how the work gets done, not whether the element was there.
The two rows at the top are the ones that matter. The ranking is computed — deterministic, recalculated from historic delivery and quality against price, proximity and customs probability, and present at the moment of decision. The weighting is human — the buyer setting what this particular requirement is optimizing for. Those two produced the result, and AI enters neither.
The split, stated plainly: the algorithm answers given what is known, who is most likely to succeed? The human answers what matters most for this requirement, and what does reality contain that the model could not yet know?
One row is worth its own sentence. Outcome write-back has no agent at all — no human, no algorithm, no AI. The service call closed or it did not, and that result governs the next ranking with nobody adjudicating. That is why the architecture held for seven years without anyone checking each decision.
What this is an excerpt from
The above is a section of a longer white paper setting out the full Human/Tech Stack™ — the derivation of each placement, the three questions to put to any agent regardless of class, and the failure modes that appear when the two load-bearing elements are reassigned.
It is available on request rather than by download. If the argument above is useful to you, the paper will be more so, and I would rather know who is reading it and why.
Request it at HPT@hansenprocurement.com.
Jon Hansen — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Related
Where AI Belongs in a Procurement Decision — and Where It Does Not
Posted on August 26, 2026
0
The Human/Tech Stack™
Jon Hansen | Procurement Insights | August 2026
THE SHORT VERSION FOR BUSY EXECUTIVES
Almost every published account of AI in procurement shows where AI goes. Very few show where it does not go, and none that I have seen explain why.
We know where AI belongs in a procurement decision and where it does not, and we can show you which is which before you commit to anything.
That comes from an architecture built in 1998 that ran for seven years — self-learning algorithms on the hardware of the period. The Human/Tech Stack™ shows what twenty-eight years of capability since has changed, and what it has not.
Ten elements of a working procurement operation. For each one: who executes it, what it delivers, and what actually moved.
Nine of the ten existed in 1998. Four are unchanged. One is genuinely new. And neither of the two elements that produced the result — the ranking and the weighting — is one that AI touches.
That is not an argument against AI. Five of the ten are the same work made cheaper, wider or continuous, and those gains are real. It is an argument about placement: the elements that decide an outcome are not the elements where capability improved.
The Human/Tech Stack™ — Hansen Models™ | Procurement Insights
How to read it
Four columns. The element. Who executes it — and every executor is named as an agent class, because a procurement operation has always run on agents: internal humans, external humans, deterministic computation, and now an AI agent. What that element delivers. And what twenty-eight years of capability improvement did to it.
The left bar and the right column measure different things, and it is worth knowing which is which. The bar says whether the core architecture is unchanged — four solid, six dashed. The right-hand column says what AI did to the work: unchanged, assists, reduces cost, expands, or new.
They are not the same axis. Six of the dashed rows existed in 1998; what changed is how the work gets done, not whether the element was there.
The two rows at the top are the ones that matter. The ranking is computed — deterministic, recalculated from historic delivery and quality against price, proximity and customs probability, and present at the moment of decision. The weighting is human — the buyer setting what this particular requirement is optimizing for. Those two produced the result, and AI enters neither.
The split, stated plainly: the algorithm answers given what is known, who is most likely to succeed? The human answers what matters most for this requirement, and what does reality contain that the model could not yet know?
One row is worth its own sentence. Outcome write-back has no agent at all — no human, no algorithm, no AI. The service call closed or it did not, and that result governs the next ranking with nobody adjudicating. That is why the architecture held for seven years without anyone checking each decision.
What this is an excerpt from
The above is a section of a longer white paper setting out the full Human/Tech Stack™ — the derivation of each placement, the three questions to put to any agent regardless of class, and the failure modes that appear when the two load-bearing elements are reassigned.
It is available on request rather than by download. If the argument above is useful to you, the paper will be more so, and I would rather know who is reading it and why.
Request it at HPT@hansenprocurement.com.
Jon Hansen — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related