Coase
27 Aug 2026 - 27 Aug 2026
- Let me see if I can reconstruct Coase's theory from my imperfect memory. It's about the differences beween interaction between agents within a firm (a command structure, more or less) and between firms (market based). Market is best and the default, firms exist because the market has transaction costs and at some scale the command structure is more efficient. I've probably screwed that up a bit
- Ask Claude
- Your reconstruction is pretty accurate. The core of Coase's 1937 paper "The Nature of the Firm" is roughly what you described:
- What you got right:
- The market is the baseline mechanism for coordinating economic activity
- Firms exist because markets have transaction costs (discovering prices, negotiating contracts, enforcing agreements)
- Within a firm, the price mechanism is replaced by authority/direction - an entrepreneur coordinates production by command rather than market exchange
- Firms grow until the marginal cost of internal coordination equals the marginal cost of using the market
- Minor refinements:
- Coase didn't quite say market is "best" normatively - more that it's the puzzle to be explained. The question he was answering was: if markets are so efficient, why do firms exist at all? Neoclassical economics at the time had no good answer
- The transaction costs he emphasized were specifically the costs of using the price mechanism, not just friction generally
- He also noted firms stop growing when internal bureaucratic costs rise to match market transaction costs - so there's a natural size limit
- The related Coase theorem (from a different paper) is probably mixed up in some people's memory with this - that's the one about externalities and bargaining, which is a separate contribution.
- Your core intuition was solid.
- How does that relate to The Network Always Beats the Castle - by Sam Schillace
- He does not refer to Coase at all, but is making a somewhat similar argument – he says, firms and organizations and command structures exist because we have not yet perfected a good agent technology (my own term, not sure how he phrases it). AI gives us one, so things can be accomplished without the rigid hierarchies of traditional command structures.
- OK Claude, read that and tell me if I got that one right, and how Coase related Ask ClaudeAsk Claude
- Your reconstruction was accurate but incomplete — you got the core right while missing the richness of his specific framing. Here's the breakdown:
- What you got right: Schillace does argue that hierarchical command structures exist because we lacked the infrastructure to do otherwise.
- His claim is that "circuit-switched organizations are the only kind that could be managed without the infrastructure that packet-switched organizations require"
- — and AI is now that missing infrastructure. So your "agent technology" gloss is fair.
- What you undersold: Your summary made it sound like a fairly simple "AI replaces hierarchy" argument. It's actually more developed than that. His central metaphor is circuit-switching vs. packet-switching in networks.
- The org chart is a circuit map — managers are "leased lines," reserved whether or not signal is flowing — and when a company needs to change, it rewires the circuit map, which we call a reorg.
- He also enumerates specific organizational functions AI can replace:
- AI can serve as the routing layer (matching work to capability without a manager in the path), the translation layer (making edge work legible without human chains), and the error-correction layer (noticing when something is blocked or off without burning human attention).
- And he's careful to note this isn't just efficiency —
- packet-switched organizations "will not just do existing organizational things more efficiently, but will do things that are not currently possible."
- The Coase connection:
- Schillace doesn't cite Coase, but the parallel is direct and illuminating. Coase asked: why do firms exist if markets are efficient? His answer: transaction costs make internal command structures cheaper at some scale. Schillace is essentially answering: what happens when those transaction costs collapse?
- He notes that experiments with flat/self-managing organizations "mostly failed" not because of culture, but because "the infrastructure simply wasn't there" — you can't run a packet-switched network without routers, protocols, and addressing.
- This maps perfectly onto Coase: the reason firms exist is precisely that the market's transaction costs (discovering, contracting, coordinating) exceeded the costs of internal hierarchy. AI, in Schillace's framing, attacks those transaction costs from the other direction — not by making hierarchy cheaper, but by making coordination without hierarchy cheap enough to work.
- In Coasian terms: AI shifts the crossover point dramatically. The firm size at which internal coordination becomes more efficient than market coordination gets much smaller — potentially down to one person with agents. The logical endpoint of Schillace's argument is that Coasian firms as we know them may largely dissolve, replaced by fluid networks that form around problems and dissolve when solved — which he does explicitly predict.
- It's a genuinely Coasian argument even without the citation.
- Your reconstruction was accurate but incomplete — you got the core right while missing the richness of his specific framing. Here's the breakdown:
- What you got right: Schillace does argue that hierarchical command structures exist because we lacked the infrastructure to do otherwise.
- His claim is that "circuit-switched organizations are the only kind that could be managed without the infrastructure that packet-switched organizations require"
- — and AI is now that missing infrastructure. So your "agent technology" gloss is fair.
- What you undersold: Your summary made it sound like a fairly simple "AI replaces hierarchy" argument. It's actually more developed than that. His central metaphor is circuit-switching vs. packet-switching in networks.
- The org chart is a circuit map — managers are "leased lines," reserved whether or not signal is flowing — and when a company needs to change, it rewires the circuit map, which we call a reorg.
- He also enumerates specific organizational functions AI can replace:
- AI can serve as the routing layer (matching work to capability without a manager in the path), the translation layer (making edge work legible without human chains), and the error-correction layer (noticing when something is blocked or off without burning human attention).
- And he's careful to note this isn't just efficiency —
- packet-switched organizations "will not just do existing organizational things more efficiently, but will do things that are not currently possible."
- The Coase connection:
- Schillace doesn't cite Coase, but the parallel is direct and illuminating. Coase asked: why do firms exist if markets are efficient? His answer: transaction costs make internal command structures cheaper at some scale. Schillace is essentially answering: what happens when those transaction costs collapse?
- He notes that experiments with flat/self-managing organizations "mostly failed" not because of culture, but because "the infrastructure simply wasn't there" — you can't run a packet-switched network without routers, protocols, and addressing.
- This maps perfectly onto Coase: the reason firms exist is precisely that the market's transaction costs (discovering, contracting, coordinating) exceeded the costs of internal hierarchy. AI, in Schillace's framing, attacks those transaction costs from the other direction — not by making hierarchy cheaper, but by making coordination without hierarchy cheap enough to work.
- In Coasian terms: AI shifts the crossover point dramatically. The firm size at which internal coordination becomes more efficient than market coordination gets much smaller — potentially down to one person with agents. The logical endpoint of Schillace's argument is that Coasian firms as we know them may largely dissolve, replaced by fluid networks that form around problems and dissolve when solved — which he does explicitly predict.
- It's a genuinely Coasian argument even without the citation.