Subtract.
The untapped science of less.

Leidy Klotz is a behavioral scientist and engineering professor at the University of Virginia. In 2021 he published a book called Subtract: The Untapped Science of Less. The Lego experiment inside it is the most quoted piece of organizational research I have referenced in the last three years.

The setup is simple. Participants are shown a small Lego structure with a roof balanced on a single corner pillar. The structure is unstable. They are asked to make it stable so the roof can hold a brick. The fix is obvious. There are two ways to do it. Add a pillar to the empty corner, or remove the roof piece that overhangs the gap. Both work.

The result.

12%

of participants improved the structure by removing a brick. The other 88% added.

Eighty-eight percent added a brick. Twelve percent removed one. The subtractive solution was faster, cheaper, and structurally equivalent. Almost nobody saw it.

Klotz ran variations of the test for years. Same answer every time. Across cultures, across professions, across age groups, the bias was consistent and measurable. Human beings are addition-default thinkers. We solve by adding. We almost never solve by removing.

When Klotz asked why, the participants had a tell. They did not consider subtraction and reject it. They did not consider subtraction at all. The option was cognitively invisible. They saw the unstable structure and their brain produced one type of solution. More.

Why this matters for the AI era.

Imagine an enterprise that has just realized AI is real. The instinct is predictable. Add an AI tool. Add a center of excellence. Add a steering committee. Add a working group. Add an internal newsletter. Add a quarterly review cycle. Add a roadmap. Add a vendor. Add a dashboard. Add a pilot. Add a second pilot.

Every addition is a brick. Every brick demands coordination. Coordination is the structure that already collapsed in the first place. The default response to an AI productivity problem is to do the exact thing that caused the problem in another shape.

Writer's 2026 enterprise AI survey found that seventy-five percent of corporate AI strategies are described as "for show" by the leaders running them. Forty-eight percent of deployments are called "a massive disappointment" by the executives who funded them. Those numbers are not a model-quality problem. They are an addition-default problem. The companies in those statistics did not lack tools. They lacked subtraction.

The framework this exposes.

I work from a simple equation called Q × A = E. Quality of capability multiplied by Acceptance of it equals Execution. It came out of GE's Change Acceleration Process, which I lived inside for twelve years. The math is multiplicative, not additive. A Q of 5 with an A of 1 produces an E of 5. A Q of 5 with an A of 6 produces an E of 30.

Most enterprises are stuck on a high Q with a collapsed A. They keep responding by raising Q. New model. New license. New tool. New vendor. Q goes up. A stays flat. E barely moves. The Coordination Tax silently accumulates, three to nine percent of gross profit, paid every year for the right to have the capability without using it.

Klotz's research is the cognitive explanation. Acceptance does not collapse because leaders chose addition over subtraction. It collapses because subtraction was not on the menu. The brain produced the only option it had practice generating.

Why it is biological, not lazy.

Klotz spent years trying to understand why the bias holds across so many populations. He landed on a set of explanations that are uncomfortable because they are out of any one leader's control.

The first is biological. Our brains evolved in environments where adding tools, allies, calories, and shelter increased survival. Subtraction had to be deliberate. Addition came pre-installed.

The second is economic. We are paid, on the whole, for things we have done. Building, shipping, launching, signing. Removing leaves a smaller paper trail. The performance system rewards the addition. Subtraction often looks, on paper, like nothing happened.

The third is cultural. We tell stories about builders, founders, and creators. We almost never tell stories about people who subtracted. The cognitive vocabulary for subtraction is thinner because the celebration of it is thinner.

Stack those three and the result is predictable. The default action under uncertainty is to add. The leader who launches a new committee in response to an AI program in trouble is not making a mistake of intelligence. They are making the default move the system installed in them long before they walked into the building.

The design move.

If the bias is cognitive, the fix is structural. You cannot wish your way out of a default. You can design a system that catches it.

That is what Phase 1 of the Velocity Framework is. The Kill List is not a productivity exercise. It is a structural countermeasure to the addition-default. For thirty days the work of the team is to name the work that should not exist. Recurring meetings nobody can defend. Reports nobody reads. Approval chains that exist because of a problem from 2019. The zombie work that survives because nobody is paid to kill it.

The output of the Kill List is two artifacts. A quantified Coordination Tax on the P&L. And a named Vital Few, the work that, if done well, makes the rest of the work matter. The Agentic Pilot in Phase 2 only deploys against the Vital Few. Everything else gets a tombstone.

Notice what changes. The team is not asked to subtract in the moment, against the cognitive grain. The team is asked to run a process whose only output is subtraction. The cognitive default is bypassed by the procedure, not by the willpower of any individual leader.

What the vanguard looks like.

The companies pulling ahead in 2026 share one trait. They subtract first. The lever is not better AI. The lever is fewer things in the way of the AI they already have.

I see it in the operating cadences. The teams that win Phase 2 do not have heavier roadmaps. They have lighter ones. They removed three quarters of what was on the calendar before they added one agent. The agent then has room to absorb work. The team has room to trust it. The org has room to feel the lift.

I see it in the leadership language. The teams that pull ahead stop using the verbs that signal addition. The roadmap conversation stops being "what should we add next quarter." It becomes "what are we ending so the next thing has room to land." That single linguistic shift, repeated in enough rooms, is the leading indicator of an operating model that has internalized subtraction.

I see it in the financial discipline. The boards that fund agentic transformation well do not approve the agent budget without a paired subtraction budget. Every dollar of new tooling is paired with a dollar of removed work, named and committed. The math is not perfect every quarter. But the principle reframes the conversation away from "how much more should we spend on AI" and toward "how much room are we making for the AI to land."

You cannot deploy agents against work that should not be done at all. The leverage point is not the agent. The leverage point is the room you make for it.

The Klotz finding is the operating instruction. The Coordination Tax is the cost of ignoring it. The Kill List is the design move that catches the bias for you. The Vital Few is what survives. That is the structural pattern. The vanguard subtracts.

The honest question.

If you sat with your senior team next week and asked them to name three things to remove from the operating model before they add one more, would the meeting be productive in fifteen minutes, or would it be silent?

I do not know your answer. I know that for most teams the answer is silent. Not because the team is bad. Because the question is rarely asked. The work of Phase 1 is to make subtraction a procedure, not an instinct, so the cognitive default stops being a strategic constraint.