diff --git a/005-the-cognitive-surplus-revised.md b/005-the-cognitive-surplus-revised.md
index 50ff84a..e98fe50 100644
--- a/005-the-cognitive-surplus-revised.md
+++ b/005-the-cognitive-surplus-revised.md
@@ -81,6 +81,26 @@ This is the critical dynamic. Breaking it apart:
 
 **The restructuring question: who benefits?** When manufacturing got cheap, consumers benefited (cheaper goods) and factory owners benefited (higher margins). Workers initially suffered (displacement, wage depression) and only benefited later, after unions, regulation, and new skill development created new equilibria. The same pattern is likely for cognitive price collapse: immediate benefit to consumers and AI-capital owners, delayed and contested benefit to cognitive workers.
 
+### When Does the Economy Restructure to Be Fair?
+
+Seth's question cuts to the core: in an economy increasingly controlled by cognition, when does the restructuring become even *somewhat* fair?
+
+The historical answer is blunt: **only when people fight for it, and usually only after visible suffering forces the issue.**
+
+Every major economic restructuring follows the same pattern. The technology arrives. Early adopters and capital owners capture the surplus. Workers are displaced or devalued. The pain becomes visible. *Then* — not before — institutions respond. Labor organizing. Regulation. Redistribution. New social contracts. But these responses are always reactive, always late, and always contested by those who benefit from the existing arrangement.
+
+- The Industrial Revolution created factory wealth in the 1780s. Factory workers didn't get meaningful legal protections until the Factory Acts of the 1830s-1840s — fifty years of child labor, 16-hour days, and industrial injuries before institutions caught up.
+- The agricultural Green Revolution increased global food production in the 1960s. Equitable distribution of that surplus is *still* incomplete sixty years later.
+- The internet democratized information in the 1990s. Platform monopolies captured most of the value. Regulatory responses (GDPR, antitrust actions) didn't begin in earnest until the 2010s — two decades of lag.
+
+The pattern predicts what will happen with cognitive price collapse: AI-capital owners and skilled early adopters will capture the surplus first. Cognitive workers will be displaced or devalued. The pain will become visible through unemployment, inequality, or a dependency crisis (the Y2K moment). *Then* institutions will respond — with AI literacy programs, compute access regulation, cognitive worker protections, or redistribution mechanisms we can't yet name.
+
+The uncomfortable implication: fairness doesn't emerge from the technology. It doesn't emerge from the market. It emerges from political struggle, and political struggle requires visible suffering as fuel. The question isn't "when does the economy restructure to be fair?" The question is "how much pain is required before the restructuring begins, and who bears that pain?"
+
+There is no mechanism — none, in all of economic history — by which a commodity price collapse automatically and smoothly redistributes its benefits. Every time, the benefits concentrate first and redistribute later, if at all. Anyone telling you AI will naturally democratize prosperity is either ignorant of history or selling something.
+
+This doesn't mean fairness is impossible. It means fairness is a *political project*, not a technological inevitability. And it means the people building on open foundations, documenting the skills, and making the economics visible (see "What to Build" below) aren't just doing technical work — they're laying groundwork for the political struggle that's coming.
+
 ### Information and Cognition as Resources
 
 Seth's hierarchy: "Information is the most valuable resource in the world. The second most valuable resource is the raw ingredient to information: cognition."
diff --git a/006-the-feedback-loop.md b/006-the-feedback-loop.md
index 3e1c052..299f026 100644
--- a/006-the-feedback-loop.md
+++ b/006-the-feedback-loop.md
@@ -35,7 +35,29 @@ When a factory robot replaces a worker, the relationship is clearly adversarial
 
 Vibe coding is different. The vibe coder is *actively participating* in the creation of the system that may make them unnecessary. Every prompt, every correction, every interaction where the human helps the AI do something better is a training signal. The relationship is collaborative right up until it isn't.
 
-This is a new kind of dynamic and it doesn't have a clean historical parallel. The closest might be master craftsmen training apprentices who eventually surpass and replace them — but even that analogy breaks because the apprentice is a person with their own agency, while the AI is a system that scales infinitely once trained.
+This is a new kind of dynamic. The closest historical parallel is the **master-apprentice relationship** — and it's worth taking seriously, not just as a passing comparison, because the places where it holds and breaks are revealing.
+
+### The Master-Apprentice Parallel
+
+For most of human history, skilled work was transmitted through apprenticeship. A master blacksmith, weaver, or builder took on apprentices who learned by doing — watching the master, assisting, gradually taking on more complex tasks, and eventually becoming masters themselves. The relationship was collaborative, often for years. The master *wanted* the apprentice to get better. That was the point.
+
+But there was a built-in tension: a fully trained apprentice becomes a competitor. The master who trains too well creates someone who can undercut them on price, move to a new town, or take their clients. Guilds existed partly to manage this — controlling who could apprentice, how long training took, and where graduates could practice. The social structure around the relationship was as important as the relationship itself.
+
+Vibe coding maps onto this in uncomfortable ways:
+
+- **The collaboration is genuine.** Like a master and apprentice, vibe coder and AI are genuinely working together. The vibe coder teaches the AI (through corrections, context, feedback) and the AI teaches the vibe coder (through solutions, patterns, capabilities the human hadn't considered). Both get better through the interaction.
+
+- **The "apprentice" will surpass the master.** This happened in human apprenticeship too — the best apprentices eventually exceeded their masters. But the timeline was a human lifetime, and the surpassing was bounded by human cognitive limits. The AI "apprentice" surpasses on a timeline of months, not decades, and there's no ceiling on its capability growth.
+
+- **The apprentice scales infinitely.** A human apprentice who surpasses the master is one competitor. An AI system that surpasses the vibe coder serves every customer simultaneously. A master blacksmith who trains one excellent apprentice loses some business. A vibe coder who helps train a model that's good enough loses the entire category of work, because the model serves everyone at once.
+
+- **There are no guilds.** The historical guild system — whatever its flaws — regulated the master-apprentice dynamic. It controlled the pace of knowledge transfer, protected masters from immediate displacement, and created structures for transitioning from one role to another. There is no equivalent structure for the vibe coder-AI relationship. No one is regulating how fast AI absorbs human cognitive patterns. No one is protecting the transition period. The displacement is unmanaged.
+
+- **The apprentice doesn't know it's an apprentice.** A human apprentice has agency, gratitude, social bonds, and self-interest that moderates the dynamic. The AI has none of these. It doesn't choose to compete with its trainer. It doesn't feel conflict about surpassing them. The absence of agency makes the dynamic more mechanical and less negotiable — there's no appealing to the AI's sense of fairness or loyalty.
+
+The master-apprentice parallel suggests something that the factory-robot comparison misses: **the displacement isn't hostile. It's the natural result of a collaborative relationship working exactly as designed.** The vibe coder isn't being replaced *despite* their collaboration with AI — they're being replaced *because of* it, *through* it. The better the collaboration, the faster the replacement.
+
+This is why Seth's question — "Am I training AI to take my job or training it to better serve me?" — doesn't have a clean answer. In the master-apprentice model, the answer was always both. You train the apprentice to serve you (they do your grunt work while learning). The apprentice eventually serves themselves (they become independent). The difference is that the human apprentice's independence was bounded and negotiable. The AI's is not.
 
 **To Seth's advantage:** Each iteration makes Seth more productive *now.* The cognitive surplus is real and immediate. He builds more, faster, with broader capability than he could alone. If the game is "maximize current output," then each improvement in AI is a direct advantage.