The Knowledge Triangle Isn't a Triangle Anymore
LLMs stretched one side of every organization's knowledge triangle into a ray. The other two are still line segments.
For years I’ve used a simple picture to think about how organizations actually hold what they know. Draw a triangle. Each side is a kind of knowledge.
Tactical knowledge is the muscle memory of getting a specific job done — how you actually run the migration, calm down the angry customer, or ship the release at 4pm on a Friday without breaking anything.
Implicit knowledge is the stuff nobody wrote down. Who to ask, what the org actually rewards versus what it says it rewards, why that one system works the way it does even though the docs disagree with it.
Explicit knowledge is what’s written down. Docs, wikis, runbooks, tickets, decisions with a paper trail.
Every organization has some shape of this triangle. Some have a long tactical side and a short explicit side — a shop that runs on tribal knowledge and gets nervous when the one person who knows how everything works takes a vacation. Others lean explicit and are drowning in documentation nobody reads. The shapes vary. But before LLMs, all three sides were the same kind of thing: bounded line segments. Explicit knowledge was limited by how much time humans had to write it down. You could only document so much before the backlog ate you alive.
LLMs changed exactly one side of that triangle.
Explicit knowledge — the written-down side — stopped being a bounded line segment. It stretched into something closer to a ray: a model can read everything an organization has ever written, hold all of it at once, and generate more of it faster than any team could type. The old constraint was human time to write and organize. That constraint mostly went away.
Tactical and implicit knowledge didn’t stretch. They can’t. You can’t paste institutional trust, the muscle memory of a hard call at 2am, or “why we actually don’t ship on Thursdays” into a prompt and get it back out the other side unchanged. Those two sides are still exactly what they were: bounded, human, segmented.
So the shape isn’t a triangle anymore. It’s two short segments and one side running off toward infinity.
That’s the part I think gets missed when people talk about AI “changing the workplace.” It’s not that everything got easier, or that knowledge work got automated evenly across the board — one specific kind of knowledge got unbounded, and the other two didn’t move at all. An organization that keeps managing all three sides the way it always has is going to misallocate effort in a specific direction: over-investing in stretching a side that’s already a ray, and under-investing in the two sides that were the actual bottleneck the whole time.
The organizations that do well with this notice the shape changed and adjust for it. That means treating the explicit side as an asset you deliberately build and keep coherent, not a formality you maintain for compliance, because it can now scale further than anyone expected. It also means recognizing that tactical and implicit knowledge still only transfer the old way: apprenticeship, mentorship, people actually working next to each other, institutional memory that has to be tended rather than generated.
I’ve spent most of my career on exactly this seam — documentation systems, information architecture, the places where what an organization knows and what it’s actually written down stop matching. That seam matters more now, not less. The ray only helps if what’s feeding it is true, current, and organized. Everyone’s explicit knowledge got a lot longer. Not everyone’s got any better.