From Knowledge to Context
An old human problem
For centuries, we have told stories, filled books, and scribbled notes on whatever was within reach to preserve what we know. We continue that work in conversations, documents, and platforms like Notion and SharePoint, trying to carry enough of our understanding forward in the hopes that we or someone else can build on these artifacts. Sometimes the reasoning travels with the record. Often, though, we are left with the result and very little of how or why it was produced. The things we encountered, the connections we made, the reasons we thought it mattered. Lost to time.
That isn't to say we are incapable of documenting that reasoning. We sometimes do, but keeping it connected to the outcome and available to whoever needs it takes deliberate, often laborious effort. Even when the notes exist, someone still has to find them and work out how the thinking recorded in them shaped the result.
The question I keep coming back to is simple. How can someone, including a later version of myself, explore what brought me to an idea and discover something in that thinking that I never saw at the time? We have plenty of frameworks, such as knowledge management, to help us define what we need to preserve, but there is nothing that I am aware of that helps us capture the reasoning behind what we produce.
What we need is the means for human experience and reasoning to become an evolving organism of collective thought, growing through the connections we make and the understanding we develop together. Giving other people and AI agents access to that connected knowledge could help us reason through a problem together, drawing on more of the thinking that shaped the approach to it. In my own graph, I am beginning to see how thinking that led to one outcome connects to thinking that led somewhere else, or never became anything finished at all.
How to make what we keep useful
Knowledge is the understanding and know-how that enable us to make judgments, solve problems, and act. The draft revision of ISO's knowledge-management standard puts it like this: “understanding that enables effective context-specific decisions and action.” That understanding can live in a person, in the practices a team develops together, or in a process someone else can follow. Some of it can be made explicit in a procedure or an explanation, while some remains tacit: the judgment we develop through experience, the things we notice, and the reasons we hesitate, which can be difficult even for us to explain. Preserving the document is one part of preserving that understanding. Helping someone use it in a different situation is where context comes in.
Imagine I am putting together a tiger team to solve a problem, and a colleague has already run one successfully. Their knowledge includes how they brought people together, the friction they encountered, and what they learned about making the team effective. I might inherit their playbook, but to use it well I need to understand what they were trying to accomplish, why they made those choices, and which constraints shaped the result. Those circumstances give me the context to judge what carries over to my team and what needs to change. If I can recover their honest reasoning, I can build on their experience while accounting for my own goals and circumstances. Otherwise, I may follow the same steps and still have to work through the problems they already learned to solve.
Even when the playbook includes the relevant ideas, decisions, and outcomes, the connections between them may never have been made explicit. It falls to me to then piece together how one thing led to another, infer why a particular approach worked, and work out what that means for the problem in front of me. That reconstruction adds another layer of effort before I can build on what someone else has already done. And if the person who wrote the playbook has left the organization or is no longer available, I lose the chance to ask about the reasoning they never wrote down or check whether the connections I am making are right.
Putting context to work
One way to erase the burden of reconstruction is to preserve the connections as part of the record. A knowledge graph lets us represent things alongside the claims we make about them. By things, I mean the people, projects, organizations, and concepts we are working with. Claims are the statements we make about them: what we observed, what we decided, what we believe, and why. A claim might record that the tiger team changed its approach because a particular market constraint made the original plan unworkable. Linking that claim to the team, the decision, and the source artifacts, such as the meeting transcript or project notes, gives someone a path through the reasoning and back to the evidence. Those connections help turn scattered records into knowledge someone can examine and use, while keeping a claim open to question rather than treating it as established fact.
Knowledge graphs have been around for years. AI now carries much of the burdensome effort of composing, managing, evaluating, and curating the graph. LLMs can also extract claims from our conversations and documents and propose how they fit together; decision models can help assess those proposals, while automated checks and human review help maintain the record as it grows. The goal is to make this an ongoing part of how we work, so that preserving useful context takes less effort and more of our experience becomes available to build on. That is the knowledge engine I want supporting my work, helping me assemble and question my understanding and letting an agent draw from it when I need help.
In The Context Inversion, I wrote about realizing that the memory I was building for my AI was giving me a way to hold onto my own thinking. Working on this piece gave me a small glimpse of what that makes possible. I had an idea of what I wanted to say, but parts of the argument were scattered across months of thinking and conversations. With an agent querying my graph, we recovered an earlier blog idea, my thoughts on human context, and a tension I had recorded between individual perspectives and agreed organizational definitions. I could recognize those threads, reconsider them, and bring them into this argument without having to remember where each one lived and why I chose to preserve it. The context was mine, and the agent helped me put it to work.
The annotations, like the one above, are small examples of earlier thinking becoming useful in a new context, recovered while we were writing, with the original wording preserved so you can see what I am building on.
Collective context
Having that context available to me is incredibly useful on its own, but what happens when other people are able to access it? If I choose to share what I learned running that tiger team, a colleague could begin with the benefit of my experience. And if they contribute what happens next, including where my approach failed or needed to change, we begin to develop a shared understanding that neither of us could have built alone. The playbook would still exist, only now it is forever linked to the reasoning that produced it. As that reasoning changes over time, so too should the playbook. That is what I mean by collective context growing from the bottom up. The people doing the work contribute what they encounter, and those contributions become connected and available to others.
To take this a step further, an enterprise armed with a mature context graph could discover unknown tensions before they erupt, eradicate duplicative work, and draw direct lines from decisions made years ago to outcomes that occurred yesterday. Shared definitions would give teams common ground, while the record could preserve differing observations and the evidence behind them.
Now, imagine a marketplace of structured thinking, where people choose to share or license the context they have carefully grown and curated over years of work. I'm willing to bet people would pay for access to the accumulated knowledge of someone who has been successful in their field. That experience could remain available long after they retire, change careers, or are no longer there to answer a question themselves. You are then able to connect your agent to that graph and immediately draw on the body of work. It brings to mind Neo in The Matrix saying, “I know kung fu.” What the agent gains is access to the expert's recorded reasoning, with its sources and limitations, so it can bring relevant experience into the problem you are working on. The value would depend on how faithfully that context captures the expert's thinking and how well it applies to your situation, but the possibility of making decades of experience accessible is extraordinary to me.
I would also be remiss if I didn't mention the implications of this technology for the masses. We are all beneficiaries, but most of us have no say in what moves a collective. That is relegated to directors, CEOs, presidents, members of Congress, generals, and the like. Communities, countries, and societies with collective graphs would let anyone follow how different groups approached a shared problem, understand the conditions behind their results, and carry that understanding into the next attempt. Over time, that could help us ask better questions, build meaningful consensus, and pursue ideas that become visible only when separate experiences meet. Agents could help us find and explore those connections, but the ambition remains human: to make more of what we learn available to one another, so we can push further together.
Why I keep coming back to this
Years ago, I was part of a Knowledge Engineering team tasked with “solving” knowledge management at the Pentagon, then the largest office building in the world. Part of that work was bridging literal silos of data so that what existed in one system could become available to someone working in another. The larger goal was to make the knowledge people accumulate available to others at the moment they needed it. I came away feeling that what I wanted to achieve was far beyond what the tools of the time could provide.
Over the course of my career, I have worked tirelessly to connect data across those boundaries. But even when the data can move, the understanding around it can remain isolated in a person's head, a team's conversations, or the history of a particular platform. Those are context silos: the information may be accessible while the reasoning needed to use it remains somewhere else. What excites me now is the possibility of carrying that context across the boundaries too, so that a change of tool, team, place, or time does not force us to reconstruct our understanding from the beginning. A shared knowledge pool could grow from those connections, giving people and their agents access to the reasoning behind what others know, including where their perspectives differ.
I've never forgotten that experience, so when I first encountered LLMs, this was the problem I wanted to solve. What amazes me now is that I can work with technologies that help me recover and connect my own thinking in ways that once required painstaking day-to-day effort. There is still considerable work between that personal experience and the collective future I have described, but I can see a path worth pursuing.
Most of us know the small strangeness of finding something we wrote years ago and meeting a stranger in our own handwriting. The words survived. The reasons did not. We treat that as a failure of discipline, as though we should have written more down. But we are active, growing minds in a rapidly changing world, with full lives on top of it. It is a wonder we write anything down at all. I would like to help build something that keeps up with us, so that the next time we meet our own old words, they feel familiar again.