Blog
The Loop, Not the Library
Marko Gargenta
Founder & CEO, PlusPlus
Intro
This spring the head of enablement at a 550-person software company walked me through the meeting she calls the most expensive one of the week. Every Monday, five or six people stand in front of the executive team and demo what they built with AI that week, including the parts that did not work. She admitted the meeting stresses her out. Then she showed me why she will never cancel it.
Here is the idea her system runs on, and it is the clearest statement of it I have encountered: collective intelligence is a loop, not a library. Knowledge compounds only when the things people build flow back into the system that helps the next person build. A library is something you fill and point people at. A loop is something that gets smarter every time someone ships.
A library is stale the day you finish it
That is not a maintenance problem. It is architecture. A library runs on a push model. Someone a step removed from the actual work writes the doc, publishes it, and hopes people come. The doc is finished the day it is written and wrong a month later, because the work kept moving and the doc did not. Companies respond by assigning someone to maintain the library, which is a race a human cannot win against an organization that changes every day.
And a library has no mechanism for the most valuable knowledge in the building, which is what people just learned by doing their jobs, to flow back in. That knowledge lives in demos, debugging sessions, and half-working prototypes, and in a library model it evaporates the moment the work ships.
The demo is the moment work becomes knowledge
A demo is a record of applied intelligence: what someone tried, in their real job, on the real system, and what actually happened, failures included.
At this company, the demo is infrastructure, not ceremony, and it shows up twice.
First, the leaderboard. To earn points for something you built, you must record a demo of it. An AI agent converts the recording into a draft playbook, getting it roughly 60% of the way there. A human expert finishes it. The playbook enters the same knowledge base their internal AI assistants answer from. Build something on Tuesday, and by the following week your experience is part of what the AI tells the next person who tries something similar.
Second, the Monday meeting. Demos in front of the executive team, every week, including what is not working. The failures are the point. A playbook tells you what works. A demo of what broke tells you where the edge is, and the edge is exactly what a new builder needs to know. The executives are not there to judge. They are there to remove blockers while the learning is still warm.
AI did not replace the experts. It made the loop affordable.
Turning a demo into a playbook used to take a technical writer days, so nobody did it, so the knowledge evaporated. Now the conversion is nearly free, and the expert's time goes where it belongs, on judgment.
This is the part most AI strategies miss. AI is only as good as the knowledge you feed it. Generic AI plus generic knowledge equals generic output. The demos are the proprietary knowledge. No foundation model knows what your engineer learned shipping last Tuesday. Only your loop can know that.
Your most valuable asset is being discarded daily
Most organizations are sitting on the record of what their people build and throwing it away. They spend real money on content libraries while the actual intelligence of the company, the applied, tested, failure-annotated kind, walks out of every demo unrecorded.
The fix is not a bigger library. It is a shorter path from doing to knowing. Make the demo the unit of knowledge. Make capturing it cheap. Let AI do the drafting and experts do the judging. Then point the whole thing back at the people doing the work.
She called it the most expensive meeting of the week. It might be the cheapest knowledge system I have ever seen.
