Andrew Grill asked a question this week that answers itself.
If AI has killed management consulting — as we are told almost daily — why did the AI companies themselves just hire the consultants? OpenAI, Anthropic, and Google have all reportedly paid the large advisory and integration firms to help land AI inside enterprises. The organizations with the most capable models on earth still need people to get those models adopted. That should tell us something.
Grill’s reading is right, and I want to take it one step further, because the answer isn’t really a story about consulting. It is a story about a pattern I have watched hold for a quarter-century.
Adoption was never a software problem.
In 1998 I led an engagement for Canada’s Department of National Defence. Delivery performance moved from 51 percent to 97.3 percent in three months — before the platform was ever selected. The improvement came from the readiness and alignment work done first; the technology was the last step, not the first. I have written some version of that sentence in every technology wave since. It was true for ERP. It was true through the pandemic, when organizations leaned on technology overnight and discovered that the pilots which stalled, stalled for human reasons, not technical ones. It is true now, for AI.
That is not a coincidence I keep noticing. It is a constant. The tooling changes — mainframe, ERP, cloud, and now agentic AI — and the requirement that the operating logic be in place first does not. I have come to call it Invariant Physics™: technologies always progress; the one thing that holds, until proven otherwise, is that the readiness has to come first. Technology-last is not technology-never. It is a sequence, and reversing it is the most reliable way to buy an expensive failure.
So what did AI actually change?
It made fluent, plausible analysis effectively infinite and nearly free. The SWOTs, the market sizing, the first-draft deck, the benchmarking that used to occupy a room of analysts for weeks — that work is now minutes. No honest practitioner should pretend otherwise.
But making one thing infinite makes its opposite scarce. When fluent analysis costs nothing, the scarce asset becomes the judgment to know which analysis is right, the trust to stake a decision on it, and the verifiable record to prove you were saying it before it was the consensus. Grill put it plainly: “the deck was never the product.” What clients were paying for was the person willing to put a judgment on the table and own it out loud — including when it turned out wrong.
That is the line I would underline, because it names both what AI cannot do and what I have built a practice around. I have kept a public, dated, never-edited archive since 2007 — not a curated highlight reel, but the thinking as it happened, the corrections left standing beside the calls that held. Being wrong in the open, on the record and timestamped, is not the embarrassment; it is the discipline. It is the whole distance between getting it right and merely looking right — and it is the only ground on which someone bets a decision they will have to live with.
This is why the AI labs are paying the consultants, and it is the part their own models cannot supply. A model will produce a confident recommendation and a flawless deck. It will not own that recommendation when it turns out wrong. It will not read the politics in a room no one will say out loud. It will not carry the accountability that turns analysis into a decision. Automating the analysis does not shrink that work — it moves all of the remaining value into it.
The consultants the AI companies are hiring are not there for the deck. AI writes the deck now. They are there for the judgment, the readiness, and the willingness to stand behind the call. That was always the job. The technology just made it, finally, unmistakable.
I have been documenting this in the open since 2007, reaching back to work that began in 1998. The record reinforces exactly the point the market is now rediscovering at scale: the operating logic comes first, and the technology — however capable — is still the last step.
This analysis draws on the Procurement Insights archive — an independent record, carrying zero vendor sponsorships, that I have published openly since 2007 and that consolidates documented client work, lectures, and writing reaching back to 1998. Every claim in it is held to the Provenance Ledger™: a verify-before-publish discipline that traces each assertion to a primary source and never quietly edits the record once it is posted. That record is the evidence base for two working lenses — Invariant Physics™, the constant that however far the technology advances, the operating logic must be in place first; and Implementation Physics™, its per-engagement application: the discipline of doing the readiness work before the platform, not after. Getting it right, rather than being right.
-30-
Related
AI Made the Analysis Free. It Made the Judgment Scarce.
Posted on July 15, 2026
0
Andrew Grill asked a question this week that answers itself.
If AI has killed management consulting — as we are told almost daily — why did the AI companies themselves just hire the consultants? OpenAI, Anthropic, and Google have all reportedly paid the large advisory and integration firms to help land AI inside enterprises. The organizations with the most capable models on earth still need people to get those models adopted. That should tell us something.
Grill’s reading is right, and I want to take it one step further, because the answer isn’t really a story about consulting. It is a story about a pattern I have watched hold for a quarter-century.
Adoption was never a software problem.
In 1998 I led an engagement for Canada’s Department of National Defence. Delivery performance moved from 51 percent to 97.3 percent in three months — before the platform was ever selected. The improvement came from the readiness and alignment work done first; the technology was the last step, not the first. I have written some version of that sentence in every technology wave since. It was true for ERP. It was true through the pandemic, when organizations leaned on technology overnight and discovered that the pilots which stalled, stalled for human reasons, not technical ones. It is true now, for AI.
That is not a coincidence I keep noticing. It is a constant. The tooling changes — mainframe, ERP, cloud, and now agentic AI — and the requirement that the operating logic be in place first does not. I have come to call it Invariant Physics™: technologies always progress; the one thing that holds, until proven otherwise, is that the readiness has to come first. Technology-last is not technology-never. It is a sequence, and reversing it is the most reliable way to buy an expensive failure.
So what did AI actually change?
It made fluent, plausible analysis effectively infinite and nearly free. The SWOTs, the market sizing, the first-draft deck, the benchmarking that used to occupy a room of analysts for weeks — that work is now minutes. No honest practitioner should pretend otherwise.
But making one thing infinite makes its opposite scarce. When fluent analysis costs nothing, the scarce asset becomes the judgment to know which analysis is right, the trust to stake a decision on it, and the verifiable record to prove you were saying it before it was the consensus. Grill put it plainly: “the deck was never the product.” What clients were paying for was the person willing to put a judgment on the table and own it out loud — including when it turned out wrong.
That is the line I would underline, because it names both what AI cannot do and what I have built a practice around. I have kept a public, dated, never-edited archive since 2007 — not a curated highlight reel, but the thinking as it happened, the corrections left standing beside the calls that held. Being wrong in the open, on the record and timestamped, is not the embarrassment; it is the discipline. It is the whole distance between getting it right and merely looking right — and it is the only ground on which someone bets a decision they will have to live with.
This is why the AI labs are paying the consultants, and it is the part their own models cannot supply. A model will produce a confident recommendation and a flawless deck. It will not own that recommendation when it turns out wrong. It will not read the politics in a room no one will say out loud. It will not carry the accountability that turns analysis into a decision. Automating the analysis does not shrink that work — it moves all of the remaining value into it.
The consultants the AI companies are hiring are not there for the deck. AI writes the deck now. They are there for the judgment, the readiness, and the willingness to stand behind the call. That was always the job. The technology just made it, finally, unmistakable.
I have been documenting this in the open since 2007, reaching back to work that began in 1998. The record reinforces exactly the point the market is now rediscovering at scale: the operating logic comes first, and the technology — however capable — is still the last step.
This analysis draws on the Procurement Insights archive — an independent record, carrying zero vendor sponsorships, that I have published openly since 2007 and that consolidates documented client work, lectures, and writing reaching back to 1998. Every claim in it is held to the Provenance Ledger™: a verify-before-publish discipline that traces each assertion to a primary source and never quietly edits the record once it is posted. That record is the evidence base for two working lenses — Invariant Physics™, the constant that however far the technology advances, the operating logic must be in place first; and Implementation Physics™, its per-engagement application: the discipline of doing the readiness work before the platform, not after. Getting it right, rather than being right.
-30-
Share this:
Like this:
Related