Blog
The Knowledge That Routes Around Your Systems
Marko Gargenta
Founder & CEO, PlusPlus
Intro
A large crypto exchange has every knowledge system money can buy. A top-tier learning platform. An enterprise wiki. AI-powered search indexed across the whole company. And a senior engineering leader there told me the most valuable learning program they run is a monthly session where engineers demo their AI workflows live, coordinated entirely by hand, by a person hired full-time to do it.
That contrast is worth sitting with, because it exposes something most organizations have not admitted. The knowledge that matters most in a company, how its own people actually do their work, routes around the official systems of record and lives in live, unrecorded, person-to-person moments. Companies do not have a knowledge storage problem. They have a knowledge capture problem.
Call it the routing problem: knowledge flows toward the lowest-friction path, and the official systems are never it.
Why doesn't the knowledge end up in the systems?
Look at each system and ask who can put knowledge into it, and the routing problem explains itself.
The learning platform holds courses. Authoring a course takes tooling, templates, and an approval chain, so the only people who author are the people whose job it is. Engineers structurally cannot contribute, so the platform fills with compliance and manager training. Useful, and completely silent on how the company builds things.
The wiki is open to everyone in theory. In practice, writing a good page is a chore done after the real work, and maintaining it is nobody's job. So it drifts. Every engineer knows the feeling of finding the page and not trusting a word of it.
The search layer is the most honest of the three. It does not create knowledge at all. It distributes whatever exists. Point brilliant search at a corpus of compliance modules and stale pages and you get brilliant retrieval of the wrong things.
So where does the real knowledge go? At the exchange, it goes to the monthly sessions. And their own numbers say so: they survey every session, and slide presentations score low while live terminal sessions, where attendees can follow along and make something work, score high. Engineers are not asking to be taught. They are asking to watch someone do the thing.
The most valuable knowledge is a verb
That survey result points at something deeper. The knowledge worth capturing is not what people know. It is what people do. A slide deck about a deployment workflow is an abstraction. Watching the engineer who built it run it in a terminal is the thing itself. This is why demos outscore presentations, why pairing teaches faster than documentation, and why the exchange's engineers pack a room every month for something no system in their stack contains.
Organizations keep trying to capture knowledge by asking experts to translate it into a document or a course. The translation is where it dies. Capture the doing directly, a recording of the demo, the session as it happened, and you skip the translation entirely.
What this has to do with your AI
Here is why this is now urgent rather than merely wasteful. Every company, the exchange included, is wiring AI into its internal knowledge. And the AI can only learn from what got captured. Feed it courseware and stale wikis and it answers like a bot trained on everyone's company. The generic corpus produces the generic assistant.
Now imagine the alternative. Those monthly sessions, recorded. Indexed. Fed into the same AI layer the whole company already queries. Suddenly the assistant answers with the workflow your own engineer demoed three weeks ago. Every session compounds. The humans in the room learn once; the corpus learns permanently, and everyone who asks afterward, human or machine, draws on it. That is the flywheel: people do, the doing gets captured, the capture feeds the intelligence, the intelligence makes the next person better at doing.
The exchange is one step from this. The hard part, engineers willing to show their work on a cadence, is already solved and already staffed. The knowledge is flowing every month. It is just evaporating in the room.
The lesson generalizes. Before buying another system for knowledge to sit in, look at where knowledge in your company is already moving, live demos, incident reviews, that one channel where the real answers appear, and put capture there. The systems of record can only serve what actually got recorded.
Knowledge that routes around your systems is not lost. It is telling you where the pipe should have been.
