Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
Howard Yu recently highlighted a remarkable shift: lower-cost AI models are winning global adoption because they are good enough, inexpensive, and improving fast. Through the lens of Clayton Christensen’s disruptive innovation theory, the trend makes perfect sense — cheaper technologies almost always capture the volume market long before they threaten the premium end.
But there is another question forming beneath the adoption numbers: not every AI interaction inside an enterprise serves the same purpose.
A marketing manager using a low-cost model to draft copy, a developer generating code on an overnight run, and a procurement team making a supplier decision worth hundreds of millions of dollars are not solving the same problem. One is optimizing personal productivity. The other is making an enterprise decision that requires governance, accountability, validation, and organizational trust. When an employee pastes a supplier contract or a spend file into a consumer chatbot and has an answer in seconds, that answer is fast, nearly free, and entirely outside anything the enterprise can see, govern, or stand behind.
Multiply that by every desk in the building and you do not have an AI strategy. You have a divide.
We have seen this movie before. It was called the spreadsheet.
Why the spreadsheet won — and never left
As I documented in December, spreadsheets keep beating AI not because they are better technology, but because humans are inherently at the wheel. You cannot use a spreadsheet without a person in the loop making judgments. The manager facing dirty data does not open the ERP; they pull the data into Excel and cleanse it themself. The spreadsheet became the individual’s own Phase 0™ — a place to assert control over the fitness of the data before trusting any system with it.
That accessibility is exactly why it could never be governed away. Decades after being declared dead, the spreadsheet is still everywhere.
Look at what grew up alongside it — APP proliferation. Even before the AI wave, the sprawl was staggering: Forrester put the average large enterprise at roughly 367 applications (2022), and Zylo put SaaS alone at around 660 (2023). In procurement specifically, most large organizations run between ten and twenty-five overlapping tools. That sprawl did not come from a plan. It came from a lack of standardization, from mergers, from function-specific need — thousands of individual, bottom-up decisions the enterprise could not see coming and could not consolidate after the fact.
That is the shadow-IT pattern in one sentence: individuals adopt what is accessible; the enterprise inherits what it cannot govern.
The new shadow tool acts
Personal AI is being adopted the exact same way — bottom-up, individually, for convenience. But there is a difference that makes the repeat more dangerous, not less.
A spreadsheet sits still. It is a file on a drive. Personal AI acts. It drafts, decides, summarizes, and recommends — and it leaves what I have called data exhaust: the trace of a judgment nobody can reconstruct. When an ungoverned tool merely stored your data, the risk was fragmentation. When an ungoverned tool makes and shapes decisions, the organization is now running on conclusions it cannot audit, from models it did not choose, on data it never meant to expose.
When a category manager pastes three supplier bids into a consumer chatbot and asks it to score them and draft the negotiation position, the time saved is real. So is the exposure: a multimillion-dollar decision now rests on a reasoning path the enterprise can neither see nor reconstruct. That is data exhaust — and it is already happening.
And the models underneath are pulling in a direction. The market is bifurcating into a cheap, good-enough commodity lane and a premium lane, and personal AI increasingly runs in the commodity lane — the least expensive, most accessible option, tuned for individual convenience rather than enterprise-grade judgment. The individual reaches for the most accessible tool. The enterprise requires the most governable one. That axis — accessibility versus governability — is the real divide, and it holds even as the cheap models keep getting better.
Why the enterprise needs the governed lane
Here is where the conversation usually goes wrong. It assumes the enterprise wants the premium lane because the model is smarter. That is not the point, and treating it as the point is the trap.
A commodity model is optimized for one person finishing one task. An enterprise model has to be optimized for something much harder: whether the organization can absorb what the model produces. A supplier recommendation is not finished when the model generates it. It has to survive procurement’s criteria, legal’s review, finance’s budget, and the operating reality on the receiving dock. The governed lane exists not because its model is more capable, but because that is where provenance, validation, and accountability live — the machinery that turns a generated answer into a decision the organization can stand behind.
The divide, in other words, is not technological. It is organizational.
And this is the part the technology conversation keeps missing: a more capable model does not close that gap. It raises the stakes of it. When the tool was mediocre, an ungoverned miss was a mediocre miss. When the tool is frontier-grade and you still skip the absorption work, the distance between what it could have delivered and what it actually delivered — the realized value you forfeit, what I call the Hansen Deflator Formula™ — gets larger, not smaller.
So how do you govern the divide?
Start with what the spreadsheet already taught us not to do.
You cannot ban it. Shadow IT proved that people route around prohibitions the moment the sanctioned path is slower than the unsanctioned one. Ban personal AI and you do not eliminate it — you lose visibility into it.
And you cannot consolidate your way out. Those 367 applications and 660 SaaS tools are the monument to the belief that one more platform will finally unify everything. It never does.
If you cannot ban it and cannot consolidate it, you govern the one thing that actually matters — the decisions the tools inform, not the tools themselves. In practice, four things:
- Classify the work, not the tool. Decide which tasks may run on personal, commodity AI and which require governed enterprise AI, with the line drawn at consequence and sensitivity, not convenience.
- Demand provenance. Any output that informs a real decision should be traceable to its source and to the model that produced it. That is the end of data exhaust.
- Keep humans at the wheel at the decision points. That is the only thing that made the spreadsheet safe in the first place.
- Make readiness the gate. Assess whether the organization can actually absorb and govern AI before scaling it. That is the purpose of Phase 0™.
There is a deeper architectural point, and I made it more than twenty years ago, long before it was fashionable: “your organization gains control of its spend environment by relinquishing centralized functional control in favor of operational efficiencies originating on the front lines.” You do not govern the divide by forcing all AI up into one central platform the front lines will route around. You govern it by giving the front lines governed capability — provenance and human judgment built in — inside a human-led, agent-based Metaprise™ model. Centralize the objective. Decentralize the architecture. That is how you close a divide instead of automating it.
Today’s Takeaway
The divide between personal AI and enterprise AI is not a technology problem, and it will not be solved by choosing the right AI. It is a governance and readiness problem wearing a technology costume. Whoever treats “personal versus enterprise AI” as “just pick the enterprise platform” will repeat the spreadsheet and app-sprawl era at higher velocity and higher stakes — because this time the shadow tool does not just hold the data. It makes the call.
The spreadsheet era fragmented data. The app era fragmented processes. The personal AI era risks fragmenting reasoning itself — and unlike the systems organizations have spent thirty years trying to consolidate, this kind does not sit in a platform. It sits inside the decisions.
The pattern does not have to repeat. But avoiding it means governing the judgment layer, not the tool — a finding that has proved itself time and again, against a record published openly since 2007 and reaching back to 1998, nearly three decades in which the tools kept changing, and the answer never did.
Keep the human at the wheel. Everything else is just faster.
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-
On the Hansen Deflator Formula™
The Hansen Deflator Formula™ measures the distance between the value an AI capability could deliver and the value an organization actually realizes from it — the share of the potential lost to inadequate readiness, governance, and absorption.
Its defining property is counterintuitive: the deflation grows as the technology improves. A more capable model raises the potential; if the organization’s ability to absorb it does not rise in step, the gap between potential and realized value widens rather than closes. Deploy a better tool into an unready organization and you forfeit more value, not less. First set out in May 2026, the measure exists to make that forfeited value visible — so it can be governed deliberately, rather than absorbed quietly as the cost of “doing AI.”
Related
The Enterprise AI Divide: Personal AI Is Becoming the New Shadow Spreadsheet
Posted on July 16, 2026
0
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
Howard Yu recently highlighted a remarkable shift: lower-cost AI models are winning global adoption because they are good enough, inexpensive, and improving fast. Through the lens of Clayton Christensen’s disruptive innovation theory, the trend makes perfect sense — cheaper technologies almost always capture the volume market long before they threaten the premium end.
But there is another question forming beneath the adoption numbers: not every AI interaction inside an enterprise serves the same purpose.
A marketing manager using a low-cost model to draft copy, a developer generating code on an overnight run, and a procurement team making a supplier decision worth hundreds of millions of dollars are not solving the same problem. One is optimizing personal productivity. The other is making an enterprise decision that requires governance, accountability, validation, and organizational trust. When an employee pastes a supplier contract or a spend file into a consumer chatbot and has an answer in seconds, that answer is fast, nearly free, and entirely outside anything the enterprise can see, govern, or stand behind.
Multiply that by every desk in the building and you do not have an AI strategy. You have a divide.
We have seen this movie before. It was called the spreadsheet.
Why the spreadsheet won — and never left
As I documented in December, spreadsheets keep beating AI not because they are better technology, but because humans are inherently at the wheel. You cannot use a spreadsheet without a person in the loop making judgments. The manager facing dirty data does not open the ERP; they pull the data into Excel and cleanse it themself. The spreadsheet became the individual’s own Phase 0™ — a place to assert control over the fitness of the data before trusting any system with it.
That accessibility is exactly why it could never be governed away. Decades after being declared dead, the spreadsheet is still everywhere.
Look at what grew up alongside it — APP proliferation. Even before the AI wave, the sprawl was staggering: Forrester put the average large enterprise at roughly 367 applications (2022), and Zylo put SaaS alone at around 660 (2023). In procurement specifically, most large organizations run between ten and twenty-five overlapping tools. That sprawl did not come from a plan. It came from a lack of standardization, from mergers, from function-specific need — thousands of individual, bottom-up decisions the enterprise could not see coming and could not consolidate after the fact.
That is the shadow-IT pattern in one sentence: individuals adopt what is accessible; the enterprise inherits what it cannot govern.
The new shadow tool acts
Personal AI is being adopted the exact same way — bottom-up, individually, for convenience. But there is a difference that makes the repeat more dangerous, not less.
A spreadsheet sits still. It is a file on a drive. Personal AI acts. It drafts, decides, summarizes, and recommends — and it leaves what I have called data exhaust: the trace of a judgment nobody can reconstruct. When an ungoverned tool merely stored your data, the risk was fragmentation. When an ungoverned tool makes and shapes decisions, the organization is now running on conclusions it cannot audit, from models it did not choose, on data it never meant to expose.
When a category manager pastes three supplier bids into a consumer chatbot and asks it to score them and draft the negotiation position, the time saved is real. So is the exposure: a multimillion-dollar decision now rests on a reasoning path the enterprise can neither see nor reconstruct. That is data exhaust — and it is already happening.
And the models underneath are pulling in a direction. The market is bifurcating into a cheap, good-enough commodity lane and a premium lane, and personal AI increasingly runs in the commodity lane — the least expensive, most accessible option, tuned for individual convenience rather than enterprise-grade judgment. The individual reaches for the most accessible tool. The enterprise requires the most governable one. That axis — accessibility versus governability — is the real divide, and it holds even as the cheap models keep getting better.
Why the enterprise needs the governed lane
Here is where the conversation usually goes wrong. It assumes the enterprise wants the premium lane because the model is smarter. That is not the point, and treating it as the point is the trap.
A commodity model is optimized for one person finishing one task. An enterprise model has to be optimized for something much harder: whether the organization can absorb what the model produces. A supplier recommendation is not finished when the model generates it. It has to survive procurement’s criteria, legal’s review, finance’s budget, and the operating reality on the receiving dock. The governed lane exists not because its model is more capable, but because that is where provenance, validation, and accountability live — the machinery that turns a generated answer into a decision the organization can stand behind.
The divide, in other words, is not technological. It is organizational.
And this is the part the technology conversation keeps missing: a more capable model does not close that gap. It raises the stakes of it. When the tool was mediocre, an ungoverned miss was a mediocre miss. When the tool is frontier-grade and you still skip the absorption work, the distance between what it could have delivered and what it actually delivered — the realized value you forfeit, what I call the Hansen Deflator Formula™ — gets larger, not smaller.
So how do you govern the divide?
Start with what the spreadsheet already taught us not to do.
You cannot ban it. Shadow IT proved that people route around prohibitions the moment the sanctioned path is slower than the unsanctioned one. Ban personal AI and you do not eliminate it — you lose visibility into it.
And you cannot consolidate your way out. Those 367 applications and 660 SaaS tools are the monument to the belief that one more platform will finally unify everything. It never does.
If you cannot ban it and cannot consolidate it, you govern the one thing that actually matters — the decisions the tools inform, not the tools themselves. In practice, four things:
There is a deeper architectural point, and I made it more than twenty years ago, long before it was fashionable: “your organization gains control of its spend environment by relinquishing centralized functional control in favor of operational efficiencies originating on the front lines.” You do not govern the divide by forcing all AI up into one central platform the front lines will route around. You govern it by giving the front lines governed capability — provenance and human judgment built in — inside a human-led, agent-based Metaprise™ model. Centralize the objective. Decentralize the architecture. That is how you close a divide instead of automating it.
Today’s Takeaway
The divide between personal AI and enterprise AI is not a technology problem, and it will not be solved by choosing the right AI. It is a governance and readiness problem wearing a technology costume. Whoever treats “personal versus enterprise AI” as “just pick the enterprise platform” will repeat the spreadsheet and app-sprawl era at higher velocity and higher stakes — because this time the shadow tool does not just hold the data. It makes the call.
The spreadsheet era fragmented data. The app era fragmented processes. The personal AI era risks fragmenting reasoning itself — and unlike the systems organizations have spent thirty years trying to consolidate, this kind does not sit in a platform. It sits inside the decisions.
The pattern does not have to repeat. But avoiding it means governing the judgment layer, not the tool — a finding that has proved itself time and again, against a record published openly since 2007 and reaching back to 1998, nearly three decades in which the tools kept changing, and the answer never did.
Keep the human at the wheel. Everything else is just faster.
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-
On the Hansen Deflator Formula™
The Hansen Deflator Formula™ measures the distance between the value an AI capability could deliver and the value an organization actually realizes from it — the share of the potential lost to inadequate readiness, governance, and absorption.
Its defining property is counterintuitive: the deflation grows as the technology improves. A more capable model raises the potential; if the organization’s ability to absorb it does not rise in step, the gap between potential and realized value widens rather than closes. Deploy a better tool into an unready organization and you forfeit more value, not less. First set out in May 2026, the measure exists to make that forfeited value visible — so it can be governed deliberately, rather than absorbed quietly as the cost of “doing AI.”
Share this:
Like this:
Related