For years, organizations invested in intranets, enterprise search, wikis, metadata, taxonomies, knowledge graphs, none of this is new. So why does it suddenly feel urgent again? Because the audience changed.

Three eras of organizational knowledge

We started organizing knowledge using folder. If you wanted to know how something worked, you needed to know where it lived:

Engineering/
    APIs/
        Identity/

Then came search, and location mattered less because you could look for what you needed. Organizations quickly discovered though, that search quality depended on knowledge quality. This realization led to years of work on metadata, ownership, ranking, freshness, and deduplication.

Eventually, the pendulum swung back. Search engines became smarter, and knowledge management systems inferred some metadata from documents. People no longer had to fill in pages of properties for every document. But the underlying work never disappeared. Organizations still had to decide which document was authoritative, which one had been superseded, and which guidance should be retired.

Now we enter the third era: reasoning. At the first glance, the difference seems subtle, but it changes everything. Where search returned documents, agents return decisions.

Search helped people find information

Search engines gave you ten documents and let you decide. When two documents contradicted each other, an experienced engineer often noticed. They asked around, knew which team had changed direction, and ignored the outdated wiki. We became remarkably good at compensating for imperfect knowledge.

Agents work differently. An agent doesn’t ask Alice which API the team actually recommends. It reasons over whatever the organization made available and adjusts its course accordingly. And finding the right information is only the beginning.

Discovery is only the beginning

Suppose you ask an agent to implement a new feature. It searches your organization’s knowledge and discovers an existing service. Great, but now it has to decide:

  • Is this service still supported?
  • Is it the preferred implementation?
  • Is there a newer replacement?
  • Does it fit this scenario?
  • Should I reuse it or build something new?

Finding information gets the agent only so far. The quality of its decision depends on the quality of your organization’s current decisions. A stale document produces more than a stale search result: it produces a stale recommendation. And the more autonomy we give agents, the more expensive those mistakes become.

The audience changed

Historically, most teams optimized their knowledge for themselves. Most questions traveled from person to person. When someone outside the team needed help, they asked and someone explained. The people who knew the answers acted as the interface. That allowed teams to get by with incomplete documentation.

Increasingly, as we use AI more and more, the first consumer of your team’s knowledge will be an agent acting on someone’s behalf. That means that from now on, every team publishes a knowledge interface.

Think of the interface as a reliable way for the rest of the organization to understand what your team offers and how to work with it. It should answer questions such as:

  • What do we own?
  • What guidance is current?
  • What should people use or stop using?
  • Which exceptions apply?
  • Who owns the decision?
  • Where should someone start?

Notice what’s missing: meeting notes, team discussions/chats, drafts, internal debates. Like a software API, a knowledge interface hides the internal discussions and exposes your team’s current decisions.

Publish decisions, not everything

Writing more documentation won’t solve this. Instead, you need to make the right things easy to discover, understand, and trust.

A team maintaining a small internal service may wonder why it should spend time curating guidance for five engineers. After all, everyone who needs to know, knows already. But say tomorrow, another team may face the same problem. When neither they nor their agent discovers the existing service, they’ll build another one, not only unnecessarily duplicating the work, but also complicating the decision tree for the rest of the organization.

AI raises the value of organizational knowledge

Knowledge management and the importance of its curation isn’t new. For years we’ve been highlighting the benefits of it to organizations. Today, as AI agents become more prevalent, that urgency increases even further. The more work we automate, the more decisions agents make without asking humans for clarification. That raises the cost of outdated guidance and contradictory recommendations as well as the cost of work that nobody outside the team can discover.

The organizations that benefit most from AI will make their current decisions easy to discover, trust, and reuse. This is necessary, because even the smartest model can’t compensate for guidance your teams never published.

In the age of agents, every team publishes an interface to its knowledge. How good is yours?