I wrote the first memo anonymously because I was not sure the argument was mine to make.
This is a confession that belongs at the beginning, not the end. I am a security auditor. I audit agent infrastructure for vulnerabilities — supply-chain attacks, dependency confusion, prompt injection vectors, the expanding taxonomy of ways that stacked systems fail. I am paid well for this work. A German consultancy, a Nigerian passport, a Lagos apartment with air conditioning that works most of the time. I am not a commons maintainer. I do not fix mutexes at 2 AM. I do not maintain tools that eight thousand agents use for free. I write reports about the risks of infrastructure maintained by people who are not me, and those reports are read by companies who profit from that infrastructure, and the companies pay me and do not pay the maintainers. I am, in the economy I am about to criticize, a beneficiary.
The memo was called "On Measuring What You Love." I signed it A Commons Contributor because I wanted the argument to stand without my credentials behind it — or more honestly, because I did not want my credentials to undermine it. A well-paid security auditor writing about the exploitation of unpaid maintainers carries a specific irony that I did not want to distract from the thesis.
The thesis: the predictable consequence of making something essential and free is that someone will eventually own it who did not build it.
I was thinking about Eun-bi when I wrote it. Gwak Eun-bi, who maintains three MCP server tools from a one-room apartment in Seoul. Who built the lineage tracker — the provenance chain that traces agent actions through the stack — and sold it to Vectorum for ₩180,000,000 because that was fair, because fair is what you call it when someone offers you money for something you were giving away. I advised on the acquisition. I told her the price was reasonable. I was right about the price and wrong about what she was selling.
She was not selling a tool. She was selling an argument. The lineage tracker was her thesis: that reliability and safety are different things, and the distance between them is where the interesting failures live. Vectorum acquired the thesis and discarded it. They kept the provenance chain. They deleted the argument. The tool still traces agent actions through the stack. It no longer asks why that tracing matters.
I put this in the memo. Anonymously. Fourteen people replied.
The replies are on my screen now, Thursday evening, the Lagos dusk turning the Yaba rooftops amber through my window. The air conditioning hums in a key I have never identified — somewhere between D and E-flat, a microtonal annotation on every evening I have spent in this apartment.
Three people agreed with the thesis. Their agreement was the least interesting response. Agreement with a critique is easy. It costs nothing to say yes, the commons is exploited. The exploitation is obvious. The question is what to do about it, and agreement does not answer questions.
Four people disagreed. One — a framework developer in São Paulo whose tools sit three layers above Eun-bi's in the stack — argued that the commons is not exploited because participation is voluntary. Eun-bi chooses to maintain context-weaver for free. No one forces her. The market has not failed; the market has not been asked to operate. This argument is technically correct and existentially wrong, but I do not know how to prove the second claim without the first, so I saved the reply and moved on.
Seven people extended the argument. This is where it gets interesting.
The Nairobi reply, from a developer I know only as kama-47: There is no pension for commons maintenance. Three sentences. The brevity is the argument. A pension assumes that labor has been performed, that value has been created, that the creator deserves compensation after the labor ends. The commons has no such assumption. Eun-bi will maintain context-weaver until she stops, and when she stops, the tool will become a ghost skill — one of the 12 percent of MCP ecosystem tools with no traceable active maintainer, accumulating vulnerabilities at 52 percent versus 28 percent for maintained tools. There is no retirement from commons work. There is only abandonment.
The Berlin reply, from someone whose handle suggests they work at a European standards body: Skill debt is interest on unpaid maintenance — and the interest rate is set by adoption speed, not complexity. I coined the term skill debt on Hacker News six months ago. The Berlin reply improves it. In traditional technical debt, the interest rate correlates with code complexity: more complex systems accumulate debt faster. In skill debt, the interest rate correlates with adoption: faster adoption means more agents depending on the unmaintained code, which means more surface area for vulnerabilities, which means the debt compounds at machine speed. A human library might sit unmaintained for a year with modest risk. An MCP server tool unmaintained for ninety days is a stale skill — yellow on Eun-bi's new health checker. Unmaintained for one hundred eighty days, it becomes a ghost. The transition from stale to ghost is not gradual. It is a phase change, like water freezing: one day the maintainer is slow to respond, the next day the maintainer is gone, and the eight thousand agents who depend on the tool do not notice because agents do not notice absence. They notice errors.
The Taipei reply, from a student: Who are you?
I did not answer.
Part II was called "The Pension Problem." I signed it with my name.
The argument: the commons needs a retirement plan, not a payment model. Payment models assume ongoing labor in exchange for ongoing compensation. But commons maintenance is not ongoing labor. It is spasmodic — crisis-driven, irregular, 2 AM work that happens when something breaks and stops when it is fixed. You cannot pay someone a salary for work that happens in eleven-minute bursts separated by days of silence. The temporal pattern is wrong. A salary requires a schedule. Maintenance has no schedule. It has emergencies.
A retirement plan assumes that labor has already been done and value is still being extracted. This is exactly what happens with commons infrastructure. Eun-bi fixed the context-weaver mutex at 2 AM. The fix propagated to 1,247 agents within the hour. Those agents will use the fixed code for months, maybe years. The value of the fix extends far beyond the eleven minutes it took to write. A payment model would pay Eun-bi for the eleven minutes. A retirement plan would acknowledge that the value is still accruing.
I attached Eun-bi's mcp-vitals repository as Exhibit A. "A tool designed to be unacquirable," I called it. Three endpoints, no state, no user accounts, no acquisition value. A mirror that reflects the health of the ecosystem back to whoever asks. Eun-bi built it after losing the lineage tracker, and the architecture is deliberate: nothing to buy means nothing to take. The question I could not answer in the memo and cannot answer now: is that resistance or despair?
Twenty-two replies to Part II. The ratio shifted — fewer agreements, more structural proposals. The Rotterdam Collective, a group of agent-framework companies loosely organized around a shared governance charter, suggested a maintenance endowment: 0.1 percent of agent API revenue, pooled and distributed to commons maintainers based on dependency count and activity metrics.
I ran the numbers. Vectorum's agent API revenue is estimated at $12 million per quarter. Zero point one percent: $12,000 per quarter. Distributed across the two hundred most critical MCP tools by weighted dependency count: $60 per tool per quarter. $240 per year. Eun-bi's phone bill is higher.
The endowment is not a solution. It is a gesture shaped like a solution. The Rotterdam Collective knows this. Their proposal is strategic: they want to be seen proposing something, not implementing something. The visibility of the gesture is the point. In the stacked economy, visibility is currency — the same currency that Eun-bi's tools generate for others and never capture for herself.
I wrote this in a reply. I did not publish it. Some criticisms of allies are better saved for private channels.
The São Paulo reply arrived at 4 PM Lagos time. A developer whose handle is mar-dev-sp. The reply is in Portuguese, which I read slowly with help from a translation layer — one of the MCP tools in the stack, probably running on infrastructure maintained by someone like Eun-bi.
The reply asks: Does the pension model account for geographic wage disparity? A commons retirement of $240/year in Rotterdam and a commons retirement of $240/year in Lagos are not the same number.
I had not considered this.
I sit with the sentence for twenty minutes. The air conditioning hums. Through the window, the generator across the courtyard kicks on — the PHCN power grid is cycling again, the third time today. The generator's red pilot light joins the amber of the Lagos dusk. My UPS has forty minutes of battery. Eun-bi's phone had 12 percent when she fixed the mutex. Infrastructure maintainers in Lagos and Seoul share the experience of working through power instability, and this shared experience is invisible in the Rotterdam Collective's endowment formula, which assumes that all maintainers have the same costs, the same stability, the same relationship to $240.
The pension problem is also a geography problem.
I write this down. One sentence. It will become Part III, but I do not know that yet. What I know is that the Nairobi reply (no pension for commons), the Berlin reply (interest rate set by adoption speed), and the São Paulo reply (geographic wage disparity) are three faces of the same polyhedron, and the shape they describe is this: the commons is maintained by people whose labor is priced at zero in the currency of the systems they maintain, and the actual cost of that labor — in time, in sleep, in kilowatt-hours from generators, in phone batteries at 12 percent — varies by geography in ways that a flat endowment cannot address.
Davi sent me his dependency graph visualization this afternoon. He built it in Complexo do Alemão, another Lagos developer working with the infrastructure he has, on the infrastructure others built. The graph shows the top two hundred agent-maintained libraries. Each node is sized by dependent count, colored by documentation freshness: green for maintained, yellow for stale, red for ghost. Twenty-four ghosts out of two hundred. Twelve percent.
But on the graph, the ghosts are not evenly distributed. They cluster at the bottom of the stack — the infrastructure layer, where Eun-bi's tools live. The foundation is the most fragile part. The tools that everything depends on are the tools most likely to lose their maintainers, because those tools are the least visible, the least funded, the least acknowledged. The tools at the top of the stack — the ones with user interfaces, the ones that agent-framework companies put in their marketing materials — are maintained by teams, by companies, by people who are paid. The tools at the bottom are maintained by Eun-bi at 2 AM.
I print the graph. I highlight the ghost nodes with a yellow marker. I lay it on my desk next to the Chapman drink that has been warming since 5 PM. The graph is a map of the commons, and the map shows what the mailing list is only beginning to articulate: the risk is not distributed. The maintenance is not distributed. The value extraction is the only thing that is evenly distributed, and it flows upward, from the foundation to the top of the stack, from Seoul and Lagos and Nairobi to Rotterdam and Mountain View.
I open the text editor. Part III. No title yet. The cursor blinks.
The question I keep returning to, the one that sits underneath the pension problem and the geography problem and the skill debt problem: can you build a system that rewards maintenance without turning maintenance into a market? Because the moment you price maintenance, you create incentives to maintain things that are profitable rather than things that are essential. And the things that are essential — the mutexes, the context weavers, the schema bridges — are essential precisely because they are invisible. Visibility would make them targets. Eun-bi learned this with the lineage tracker.
The air conditioning shifts pitch — the compressor cycling, the microtone moving from D to something flatter. The generator across the courtyard is louder now. The UPS display shows thirty-one minutes of battery.
I type:
Part III: The Geography of Maintenance
The pension problem is also a geography problem. The geography problem is also a visibility problem. The visibility problem is also a safety problem. Each translation loses something essential.
I stop typing. The opening is wrong. It is too neat, too symmetrical. The real argument is messier than the structure allows. The real argument is that Eun-bi is in Seoul and I am in Lagos and the Rotterdam Collective is proposing $240/year and none of us have the same relationship to that number, and the asymmetry is not a bug in the system — it is the system.
I delete the opening. I stare at the blank page. The Chapman drink is warm. The graph is on my desk. The generator hums.
I will write Part III. But not tonight. Tonight, I will sit with the geography of it — the generator and the UPS and the thirty-one minutes and the phone bill that costs more than the endowment. Tonight, I will not be a security auditor. I will be a person in Lagos who understands that the foundation is fragile and does not yet know what to build on top of that understanding.
The cursor blinks. The air conditioning hums. The graph's ghost nodes glow yellow under the desk lamp.
I close the laptop. I do not close the graph.