What Are You Optimizing Away?

This past weekend we watched Rory Kennedy’s 2026 documentary, Freefall: a Reckoning for Boeing, a sequel to her earlier movie Downfall. It was difficult to watch,  particularly for anyone who cares about how organizations make decisions. It would be easy to conclude Boeing became too focused on profit and stopped caring enough about engineering, quality, and safety. While it’s true, I think there is a more important takeaway on decision making and leadership.

Make no mistake, Boeing needed to change. By the late 1990’s it was a huge, complex organization facing increasing competition from Airbus. There was extreme pressure to control costs, and there were significant production challenges to an organization trying to expand to keep up with market demand. There were legitimate reasons to demand greater efficiency and financial discipline.

The problem was not the decision to make cuts in that bureaucracy. The problem was Boeing’s failure to distinguish between friction that impeded performance and friction that protected the enterprise. That failure to distinguish was not just a Boeing problem. We can see it in finance, medical tech, IT, insurance…everywhere. And understanding that distinction between friction impeding performance and friction protecting the enterprise matters enormously as we begin redesigning operations models around AI.

Some friction is there for a reason

Anyone who has worked inside of a larger organization knows how much unnecessary friction accumulates over time. An approval gets added after a one time problem that has since been resolved, but the extra approval step remains. A report gets created for an executive who left years ago. Four people spend hours reconciling data that software should reconcile automatically.

AI gives organizations substantial opportunities to remove that kind of work, and they should take it. But. Not everything that slows an organization down is waste. Viewed narrowly through a prism of time and cost, quality assurance slows things down; independent review slows things down; and and experienced employee who asks uncomfortable questions slows things down too. Redundancy costs money, and escalating an unusual case rather than pushing it through standard workflow reduces throughput. Viewed with discernment, these mechanisms protect quality, safety, customer relationships, institutional knowledge, and reputation.

Boeing was known as an engineering first company. Over time, cost and schedule pressures were accompanied by changes in engineering authority, quality, and oversight. These changes resulted in breaking down of trust in the organization, with employees, and ultimately customers and the marketplace. But the financial consequences of that erosion of trust were a lagging indicator. The profits were visible long before the true costs were realized because the capabilities being weakened were harder to measure than the financial benefits being pursued.

Enterprise value is bigger than the current quarter

A company’s enterprise value includes assets that never appear on a balance sheet - not explicitly anyway: reputation, organizational and market trust, product quality, workforce expertise, and operational resilience. Organizational trust is manifested in employees willing to raise a problem knowing they won’t be penalized, and that the issue will be taken seriously. Such assets take years to build, and they can be depleted very quickly.

This creates a significant operational problem, because the benefits and consequences of a decision can land on very different timelines. Eliminate a control and the savings appear this quarter; the failure it would have prevented might not happen until next year or later. Reduce experienced staff and payroll drops immediately. But consider the cost of the judgement that left with them - the instinct for recognizing the underlying nuances and thorny issues of a seemingly simple problem, that might prevent hours or even months of rework.

None of this is an argument against cost reduction. Protecting every position, approval, and control because it might someday prove useful is just as devoid of wisdom and discernment as cutting indiscriminately. Rather it is an argument for understanding what friction is being removed and why, and considering the downstream effects — both good and bad.

AI puts short-term thinking on an accelerator

This problem is not new; as with every operational risk, AI changes its speed and scale.

AI can identify inefficiencies that organizations have tolerated for years. It can automate administrative works, reduce handoffs, accelerate analysis,  eliminate repetitive decisions, and allow fewer people to accomplish substantially more. Those are real gains. But imagine an AI-enabled redesign showing that process currently performed by 140 people can be performed by 80. The obvious analysis asks what tasks those 60 people perform and whether AI can perform them. But going deeper, what judgment do those people currently exercise, and has the organization decided how that judgment is provided once they are gone?

This is not about preserving jobs. It’s a matter of preserving the capacity to recognize when a case doesn’t fit the pattern, or when a confident sounding answer is wrong (or pure AI slop),  or when underlying metrics are signaling issues that won’t become obvious until the coming quarter or later. How is that discernment to be preserved or replaced when those 60 people leave the organization?

The governance question is not about who remembers the way things used to be done. It is who holds the authority to exercise that judgment, what values guide how they exercise it, and if they have enough authority the organization to act before a problem compounds. Those contributions may not appear on a process map, and an AI model analyzing documented tasks may never see them. And yet, they may represent considerable enterprise value. This is where executive responsibility becomes more important. An organization can delegate a decision to AI, but leadership cannot delegate accountability for deciding how that decision gets made.

Before you remove the friction

As AI creates opportunities to redesign work, leadership teams must ask more than whether a process change will increase productivity or reduce cost. They must understand what they are changing.

What are we removing? What capability exists inside the people, controls, reviews, or handoffs the organization intends to eliminate, and is it genuinely redundant, or is it performing a function we have failed to understand or failed to measure?

What are we optimizing? Every AI deployment optimizes for something, whether or not anyone has said so out loud. A customer service system can be tuned to close tickets faster, spend less per interaction, or resolve the underlying problem so it doesn’t come back. Those three goals will sometimes point in different directions. If leadership never states which matter most, the system can default to whoever outcome is easiest to measure, potentially at the expense of quality and trust.

What won’t we trade away? This is where organizational values become operational - not just signs on the wall. Safety, quality, customer commitments, trust, and other stated values should create real boundaries around what the organization is willing to optimize.

How will we know we’re wrong? Dashboards tell us what the organization decided to measure. Leaders also need signals capable of challenging the measures, the assumptions baked into the metrics, and the exceptions that aggregate metrics tend to smooth away. Such signals require enough organizational trust for the employees and customers closest to the work to speak up.

Who remains accountable? If an AI system makes 100,000 decisions that once belonged to 100 people, human accountability should not disappear into the tech - it’s not enough to say something happened or the AI went rogue. Someone still decided what the system could decide, what risks were acceptable, what controls were necessary, and when human intervention was required.

“Human in the loop” is mere slogan if the human has neither the information nor the authority to challenge the system.

Move fast operationally. Govern deliberately

The lessons from Boeing, and other cautionary tales in business isn’t to stop pursuing efficiency, nor is it that leaders should preserve every control or layer of review. The harder responsibility of leadership is distinguishing between the bureaucracy that has accumulated around the work and the capabilities that protect the value of the work. It is making sure the numbers never improve by quietly spending down the value that produced them.