Writing the Specs

Every outsourcing decision changes more than who does the work. It can also change where the knowledge behind that work lives.

In 2001, Boeing structures engineer John Hart-Smith wrote an internal paper warning about the risks of extensive outsourcing. His concern went beyond cost. If Boeing outsourced too much of the work, it risked losing the technical knowledge required to specify, oversee, and evaluate what its suppliers produced.

That distinction is worth sitting with. The risk wasn’t simply losing the ability to build something. It was losing enough understanding of the work that you could no longer say precisely what you needed or confidently judge what came back.

Boeing gives us a way to see why that matters.

In 2005, Boeing sold its Wichita commercial aircraft operation, along with facilities in Oklahoma, to an investment group that formed Spirit AeroSystems. Spirit then became Boeing’s exclusive supplier for much of what Wichita had produced internally. At the same time, Boeing was developing the 787 around an unusually extensive global supplier model, with major sections of the aircraft produced by suppliers around the world and Boeing retaining responsibility for design and final integration.

The 787 program exposed significant problems in managing and integrating work distributed across that network. Suppliers delivered late or incomplete work, and Boeing sent engineers to suppliers to resolve problems and bring production back on track.

What the experience made visible was how much knowledge can live inside the doing.

The drawings and specifications matter, but so do the exceptions, the mistakes, the small decisions, and the repeated experience of seeing what happens when something changes. Some knowledge lives in people who have been close enough to the work for long enough to recognize when something does not fit.

The years that followed brought consequences well beyond schedule and cost. Two 737 MAX crashes in 2018 and 2019 killed 346 people, and the FAA revoked the aircraft’s airworthiness certificate. In January 2024, a door plug separated from a 737 MAX 9 in flight after being incorrectly installed at the factory.

Different aircraft, different causes, different investigations. What they share is a pattern of problems becoming visible to someone outside the company before they were caught inside it, and when a regulator revokes a certificate, an external body has taken over a judgment the organization could no longer supply for itself.

Nearly two decades after selling the Wichita operation, Boeing announced in 2024 that it would bring Spirit AeroSystems back inside the company in a transaction valued at about $8.3 billion including net debt. Boeing described the move in terms of production alignment, safety, quality, and workforce stability. The acquisition was completed in December 2025.

There were many factors behind the problems Boeing experienced and many reasons for bringing Spirit back inside. Outsourcing does not explain everything that happened at Boeing.

And outsourcing itself isn’t the problem. Organizations have used it successfully for decades, often gaining access to capabilities that are deeper or more efficient than they need to maintain themselves.

The harder question is what the organization still needs to be capable of once the work moves.

That is what makes Hart-Smith’s warning so useful. Boeing did not need to retain the capability to make every component itself. It did need enough capability to specify the work, integrate it, and judge what came back.

AI makes that distinction even more important.

When work moves to another company, the capability to do it continues developing somewhere. People are still encountering the exceptions, making mistakes, recognizing patterns, and learning through repetition. The organization may decide it no longer needs to own that capability itself.

When AI does more of the work, the mechanism changes.

Human capability can still develop, but through a different kind of experience. A person working closely with AI is evaluating what comes back, questioning it, adding context, recognizing patterns, challenging outputs, and seeing what changes the outcome. They may no longer be doing every part of the work themselves, but that does not mean they have stopped learning.

It means we have to think differently about what they are learning and how.

For years, organizations did not have to think very hard about some of this. People developed judgment because they spent years doing the work. The work itself was the teacher.

Now we have to be much more deliberate.

As AI takes on more execution, we don’t need to preserve every capability that execution used to require. But we do need to know which capabilities still need to exist inside the organization, and create the conditions for them to continue developing.

Writing the specs requires capability, and so does knowing whether what comes back actually meets them.

The work can move.
The harder decision is knowing what capability has to stay.