In the last post, I introduced the term Deep Stack Dividends … the categories of high-value technical work that have always existed, have always been needed, and have never been economical to fund at scale under the old organizational model.
This is the post that unpacks what that actually means for a leader trying to make a defensible capital argument in the current environment.
The investment asymmetry hiding in plain sight
A relative recently completed mandatory AI training at a major corporation. One day. The slide deck covered what a token is and how to write a persona prompt. No why. No dangers. No limitations. Nothing about when human judgment still has to lead.
Billions invested in the infrastructure. One day invested in the people using it.
That ratio isn’t an accident. It’s a symptom. Organizations have been optimizing for tool acquisition and deployment speed while systematically underinvesting in the human capability required to govern what they’ve deployed. The result is what one analyst recently described as a slot machine: you swapped a known cost for a mystery bill. The meter is running. The outcomes are unpredictable.
This is the investment asymmetry at the center of most AI transformations right now. And it points directly at the opportunity.
What the data actually shows
Gartner recently studied 350 global business executives at companies with over a billion dollars in annual revenue. Eighty percent of companies that piloted AI or autonomous technology carried out workforce reductions. Of those, there was no correlation between the layoffs and higher ROI.
Read that again. The dominant playbook … pilot AI, reduce headcount, capture efficiency gains … is not producing the returns it promises. The companies actually seeing high ROI were doing something different. They kept their people and used AI to make them more productive.
This isn’t a practitioner argument. This is an empirical finding about capital allocation strategy. The amplification model outperforms the replacement model. The organizations treating their practitioners as appreciating assets rather than reducible headcount are winning the return on investment competition.
Anthony Calleo, a former Disney executive who now works on organizational friction, arrived at the same conclusion independently: “The people who create extraordinary outcomes with AI later are usually the same people who spent years developing judgment, pattern recognition, context, and craft before the tools arrived. That capability doesn’t appear out of nowhere.”
He added the risk framing directly: “The real risk is what happens when organizations optimize for immediate efficiency while unintentionally removing the very friction that develops people in the first place.”
Two independent data points … empirical and observational … converging on the same mechanism. The friction that looks like inefficiency is often the developmental substrate that makes high performance possible later. Remove it in the name of optimization and you’ve quietly liquidated an appreciating asset.
What’s actually on the table
Deep Stack Dividends are the yield categories that become economically viable when AI absorbs the repetitive layer and frees practitioners to do the work that compounds. They’ve always existed as line items that lost the budget conversation every quarter. Under a High-Yield Governance model, they’re the return on the investment.
Documentation Archaeology is the recovery of buried system knowledge before the people who carry it leave. Every mature enterprise system has critical operational logic that exists nowhere in writing … only in the heads of practitioners who’ve been running it for years. When those practitioners leave, the knowledge leaves with them. The organization then spends years and sometimes decades rediscovering what it once knew. AI-augmented practitioners can now surface, structure, and preserve that knowledge at a scale that was previously uneconomical. The asset appreciates. The risk declines.
Root Cause Analysis as a Service is finally funding the defect categories that preventative testing never reaches. Most organizations know their recurring failure patterns. They’ve known them for years. The reason those patterns persist isn’t ignorance … it’s economics. Deep root cause analysis is expensive in practitioner time, and that time has historically been consumed by the regression layer. When AI runs the regression layer, the time recaptured can fund the investigation that actually closes the loop. Defect recurrence drops. The system gets more stable. That stability has a measurable balance sheet value.
AI Validation is maintaining craft-level oversight of AI outputs … testing the tester. This is the category that most organizations are currently leaving entirely unaddressed. AI systems generate outputs at a speed and volume that makes human review feel impossible under the old model. Under High-Yield Governance, practitioners aren’t reviewing every output … they’re governing the system that produces outputs, calibrating confidence thresholds, flagging drift, and maintaining the oversight quality that the Wiles et al. research confirms is already degrading in organizations that have encoded AI into their hierarchy without addressing this gap.
Exploratory Augmentation is deep-stack risk discovery at a scale solo practitioners could never achieve manually. The exploratory testing discipline has always been the highest-signal activity in a quality practice … and the first to be cut when capacity gets constrained. AI-augmented exploratory work changes that equation. A practitioner who previously had capacity for two or three deep exploratory sessions per sprint can now cover territory that previously required a team. The risk surface gets mapped. The organization stops being surprised by the failures it should have anticipated.
Why this never got funded before
The honest answer is that these categories lost the budget conversation because they were competing against headcount justifications in a positional hierarchy. The org chart evaluates work by the position that performs it. Deep Stack work doesn’t fit cleanly into a position … it’s investigative, variable, judgment-intensive, and difficult to forecast. It looks like overhead in a headcount model. It looks like yield in a portfolio model.
That’s the reframe. Not “can we afford to fund root cause analysis” but “what is the cost of not funding it” … measured in recurring defect rates, system brittleness, undocumented risk, and the compounding liability of organizational knowledge that exists only in people who will eventually leave.
A CFO presented with that framing isn’t looking at a cost center. They’re looking at an unhedged risk position that has a known mitigation with a calculable return.
The compounding effect
Here’s what makes Deep Stack Dividends structurally different from traditional efficiency arguments: they compound.
Documentation Archaeology produces artifacts that make every future practitioner more effective. Root Cause Analysis produces system improvements that reduce future defect rates. AI Validation produces governance frameworks that scale. Exploratory Augmentation produces risk maps that inform architectural decisions.
Each category generates not just immediate value but durable organizational capital … the kind that shows up on a balance sheet as an appreciating asset rather than a consumed resource.
The organizations that fund this work now will have systems that are better documented, more stable, better governed, and more thoroughly risk-mapped than their competitors … not because they spent more, but because they allocated differently.
Appreciate, Don’t Liquidate.
Rick Cavallaro is President of RJCadvisors Inc. and the author of The New Beautiful Testing. The full framework is at qameditations.com.