The Past Needs an Editor

In early 2016, I was editor in chief on the global brand team at Hewlett Packard Enterprise. HPE had just separated from HP, making it a $50 billion startup with decades of history but no established brand, editorial voice, or point of view of its own.

Our first major task was building HPE.com, beginning with an inventory of the enterprise technology pages scattered across HP.com.

Nobody could tell us how large the site was. Estimates ranged from seven million to nine million pages.

Decades of acquisitions had brought new businesses, products, regions, and publishing systems into HP. Under a highly decentralized model, teams had been free to create pages as they saw fit.

The result was a digital landscape containing valuable technical knowledge alongside expired promotions, abandoned blogs, obsolete product pages, forgotten reports, and years of disconnected corporate thinking.

The site recalled Jorge Luis Borges’ short story “The Library of Babel,” which evokes an imagined universe containing every possible book. Somewhere within it lies every truth—but also every falsehood and meaningless variation. Total information makes understanding nearly impossible.

HP.com was not infinite, although at nine million pages it sometimes felt that way. The company’s knowledge was in there. So was everything that obscured it.

Building HPE.com required more than migrating content. We had to decide what the new company should remember. We preserved the history that still mattered, discarded what no longer did, and positioned HPE as a company rooted in Hewlett-Packard’s technical legacy but focused on the future of computing.

HP.com was not infinite, although at nine million pages it sometimes felt that way.

That experience taught me that organizations do not forget only by losing information. They also forget when information becomes too scattered, voluminous, and incoherent to guide the next decision.

An archive is accumulation. Memory is selection. Judgment determines what the organization carries forward.

That distinction matters because AI is turning every organization into its own Library of Babel. A model can search millions of documents in seconds. But it cannot independently determine which contain durable knowledge, which have been superseded, or which lessons should constrain a decision today. Without context, provenance, and judgment, AI does not transform an archive into memory. It gives everything—including errors, contradictions, and obsolete assumptions—a new voice.

HPE shows what happens when knowledge is buried under accumulation. But organizations can forget in a more dangerous way: they can record the right lesson, reform around it, and still fail to make it govern the next consequential decision. Few institutions have confronted that failure more starkly than NASA.

Lessons unlearned

On January 28, 1986, the Space Shuttle Challenger broke apart 73 seconds after launch, killing all seven crew members. The immediate cause was the failure of O-ring seals in unusually cold weather. But the relevant knowledge was not missing. Engineers at Morton Thiokol had expressed concern about the cold and recommended delaying the launch.

The warning existed. It did not acquire enough institutional force to stop the launch.

After Challenger, NASA changed management structures, strengthened training, and created new ways to capture what its people had learned. In 1994, it launched the online Lessons Learned Information System to preserve knowledge across programs and generations.

Then, in 2003, Columbia disintegrated during reentry, killing another seven astronauts.

The Columbia Accident Investigation Board found disturbing parallels. Before Challenger, repeated O-ring erosion had gradually been accepted as manageable. Before Columbia, repeated foam strikes had undergone the same normalization. Because earlier missions had survived, evidence of danger was interpreted as evidence that the system was safe.

Seventeen years and 87 missions separated the accidents. The board asked: “How could the lessons of Challenger have been forgotten so quickly?”

An archive is accumulation. Memory is selection. Judgment determines what the organization carries forward.

NASA had preserved the reports, implemented reforms, and institutionalized the lessons. What faded was their power to shape behavior when schedules, hierarchy, and apparent success pushed in the opposite direction. The board found that NASA still lacked a systematic process for identifying dangerous trends. Problem-reporting systems were overloaded or underused.

The pattern continued. A 2012 inspector general audit found that NASA’s lessons database had become marginalized. Twelve of the 28 project managers surveyed had never used it. Only six considered it useful. Managers described it as outdated, difficult to search, and largely irrelevant to the decisions they faced.

NASA possessed the information. The challenge was getting the right lesson into the right decision before history repeated itself.

Memory needs authority

Modern NASA has built a broader answer.

After Columbia, it established the NASA Engineering and Safety Center to independently test and assess high-risk projects. NASA also separated technical authority from program management, creating checks and balances between those delivering a mission and those judging whether it is safe.

NASA’s formal dissent process allows unresolved objections to bypass the chain of command whose schedule or budget they might threaten, moving upward even to the NASA administrator. Both the disagreement and decision must be documented.

NASA has also built a knowledge network around its repositories. Its chief knowledge officer and counterparts across NASA’s centers oversee oral histories, mentoring, after-action reviews, continuity planning, and knowledge transfer during employee transitions. The objective is to preserve not only technical findings but the context required to use them.

An organization remembers only when past experience changes present behavior.

These structures do not guarantee learning. NASA’s 2025 Aerospace Safety Advisory Panel report warned that workforce losses, budget pressure, inconsistent standards, and increasingly complex missions continue to threaten technical authority. Independence can easily become advisory rather than authoritative.

That may be NASA’s deepest lesson. Institutional memory is not something an organization builds once. It must be continually renewed through structures that give experience, expertise, and dissent enough power to change what the organization does.

An organization remembers only when past experience changes present behavior.

Retrieval vs. remembering

AI can improve access to institutional memory. It cannot supply the judgment the archive lacks.

In a 2023 study of content reuse, Microsoft researchers identified one reason why. Organizational knowledge derives meaning partly from its provenance: who produced it, why it was created, and how it relates to the wider work of the organization. When AI detaches content from that context, it can surface the words without revealing how they should be interpreted—or whether the organization still stands behind them.

A 2025 study of retrieval-augmented AI systems demonstrated a related danger. Outdated information reduced accuracy and misled models even when current information was also available. Retrieving more material did not help the system distinguish the authoritative source from the obsolete one.

That distinction matters inside any large organization, where archives contain competing drafts, superseded strategies, abandoned claims, and decisions made under conditions that no longer exist. Making all of it instantly accessible does not produce institutional memory. It makes editorial judgment more important.

What organizations should do

Capture decisions, not just documents. Preserve the evidence, alternatives, assumptions, and judgment behind important conclusions.

Build memory into the work. Capture lessons while the people and context are still present.

Preserve provenance. Make authorship, sources, status, and revision history visible to humans and machines.

Connect knowledge to people. Reveal who understands the issue, not merely which file mentions it.

Forget deliberately. Retire obsolete claims, superseded evidence, and abandoned material.

The AI era will give organizations unprecedented access to their accumulated information. That does not mean they will understand it.

AI can open every drawer in the archive, but it can't decide what the organization should carry forward. That remains an act of institutional judgment.