Your senior engineer just rewrote the same Claude prompt for the fourth time this week. Your PM is pasting context into a fresh chat because nobody told her the staff engineer already solved that exact pattern in March. Your designer is one-shotting components with a prompt that took her two months to refine — and it lives in a Notion doc three people can find.
That is knowledge debt. Conservatively, a 12-person product team is burning 8-15 hours a week re-prompting from scratch. At a $150 blended rate, that is $93,600 to $175,500 a year set on fire. Not on bad work. On work that already existed inside the building.
Now ask the better question: what happens when that same team structures its prompts, context files, and agent instructions into a shared, versioned library that compounds every week? The output curve bends. The new hire ships in week two. The senior people stop being human caches.
I'm Matt Cretzman, and I have been building this stack — for myself, for clients, and inside the Knowledge Delivery System I run at Skill Refinery. The debate happening in Lenny's community right now about how teams share Claude Code context is not a tooling debate. It is the early signal that prompt and context libraries are becoming the real org chart.
The Org Chart Is Dead. The Context Graph Is Alive.
For twenty years the org chart told you who knew what. You routed questions to people. People were the index.
That model is breaking in real time. When a junior engineer with the right CLAUDE.md, the right agent definitions, and the right retrieved context can ship work that used to require a principal engineer's brain, the question is no longer "who do I ask." The question is "whose context am I loading."
The team's expertise is no longer trapped in heads, Slack DMs, and one-off Loom videos. It is either structured into shared context — versioned, tagged, retrievable — or it is leaking. There is no third option.
This is what I mean by the expert economy. Knowledge is being extracted with or without the expert's consent. The only variable is the terms.
The Three Extraction Economies, Applied to Your Team
There are three ways your team's expertise gets extracted right now. Pick honestly.
Their terms, version one: the model ate it. Every prompt your team sends to a frontier API without a structured retention strategy is a small donation. The expertise that made the prompt good — the constraints, the taste, the hard-won "do not do this" — gets used to shape the response, then evaporates from your side of the ledger. The model company keeps the aggregate signal. You keep the output for one session.
Their terms, version two: the hourly dump. Mercor and platforms like it now pay your best people $80-$200 an hour to dump their IP into someone else's training set. It feels like a side gig. It is an extraction economy. Your senior staff engineer is being paid retail to give wholesale access to the exact judgment you hired her for.
Your terms: expert-owned context. The team builds, owns, and versions the context library. The prompts, the agent definitions, the eval sets, the retrieval corpora. The cards stay on your side of the table. The compounding stays on your side of the table.
Most teams are running one and two by default. Three takes intention.
What a Shared Claude Context Layer Actually Looks Like
Let me get concrete. Here is the stack I have built and watched work.
Layer 1: The CLAUDE.md spine. Every repo and every project has a root context file. Architecture, conventions, the "do not touch" list, the test philosophy, the deploy pattern. Not aspirational. Actual. Updated when reality changes.
Layer 2: Role-shaped agents. Reviewer agent, refactor agent, migration agent, copy agent, research agent. Each one is a markdown file with a defined job, defined inputs, defined output contract. Versioned in git. PR'd like code. When the senior engineer learns a new pattern, she updates the reviewer agent. Now everyone is reviewed by her ghost.
Layer 3: The prompt library. Tagged, searchable, indexed. Not a Notion graveyard. A retrieval system. When the PM needs to write a launch brief, she does not start from a blank chat. She loads the brief-writer context, pulls three prior briefs as examples, and ships in twenty minutes.
Layer 4: Eval and feedback. When a prompt fails, that failure becomes a test case. The team's mistakes turn into the team's moat. This is the layer most people skip. It is the layer that compounds.
Layer 5: Distribution rules. Who can read which context. Who can write to which context. What context never leaves the building. The same instincts you have about source code, applied to the upstream artifact that produces source code.
This is the Knowledge Delivery System pattern. Capture once. Structure once. Deliver infinitely. It is what I built Skill Refinery to do for experts who sell what they know, and it is the exact same pattern that turns a 12-person team into a compounding asset.
The Knowledge Debt Math
Let me show you the gap.
Team A has no shared context layer. Every engineer prompts from scratch. Average prompt iteration to get a usable result: 6 turns. Average prompts per engineer per day: 25. That is 150 turns of context reconstruction per engineer per day. Across 12 people, 1,800 turns a day spent re-explaining things the team already knows.
Team B has a structured context layer. Average iteration to usable result: 2 turns. Same 25 prompts. 50 turns. Across 12 people, 600 turns a day.
Team B is not 3x faster. Team B is compounding. Every good output becomes a new template. Every failure becomes a new eval. Every new hire inherits the full library on day one, not month nine.
By month twelve, Team A is paying a knowledge tax that grows every quarter. Team B is sitting on an asset that grows every quarter. Same headcount. Same tools. Different terms.
Why This Is an Org Chart, Not a Wiki
A wiki is a place people go to read. A context library is a place models go to act. The difference is everything.
When your context library is the thing routing work, defining quality, and shaping output, it is doing what the org chart used to do. It is saying who is in charge of what. The senior engineer's reviewer agent is reviewing every PR. The head of design's component-naming context is naming every new component. The founder's voice file is writing every launch email.
Seniority becomes structural instead of synchronous. The principal engineer does not need to be in the room. Her context is in the room.
This is what stewardship of expertise looks like in 2026. You took years to learn what you know. The gift is real. The calling is to structure it so it outlives the next meeting, the next sprint, the next role change. Knowledge that does not get structured does not get stewarded. It gets extracted.
The Move
If you run a team, here is the work this quarter. Not next quarter.
One. Audit your prompt surface. How many Claude, Cursor, and ChatGPT sessions happen across the team in a week. Where does the output go. Where does the input go. Most teams cannot answer this. That is the problem.
Two. Pick one workflow and structure it. Code review, brief writing, customer reply drafting, whichever bleeds the most. Build the context file. Build the agent. Version it. Make one person the owner.
Three. Make context contributions a promotion criterion. The senior people on your team are senior because of what they know. If what they know is not in the library, they are not actually senior to the system. They are senior to a room they happen to be in.
Four. Decide your terms. Which context is shared internally. Which context is your competitive moat and never leaves. Which context, if any, you license externally. These are real decisions now, the same way IP decisions were real decisions in 2005.
I wrote a book about this — On Whose Terms: The New Expert Economy and the Fight for What You Know. The Claude context fight inside your team is the small version of the fight every expert is in right now. Same physics, different scale. Grab a copy.
The teams that win the next five years will not be the teams with the best models. Everyone gets the same models. The winners will be the teams that turned their expertise into shared, versioned, owned context — and refused to dump that context into someone else's economy.
The org chart is being rewritten in markdown. Either you are writing it, or your tools are writing it for you. More at mattcretzman.com.
Keep Building,
— Matt