The Company That Doesn't Exist — Yet
AI doesn't simply shrink or flatten organizations — because it cuts transaction costs on both the market side and the internal-coordination side of Coase's classic firm-boundary problem, it may enable both radically smaller firms and radically larger ones, while concentrating what remains scarce (entrepreneurial judgment) even more tightly at the top.
Summary
Responding to Jack Dorsey and Roelof Botha's 'From Hierarchy to Intelligence' essay on Block's AI-mediated reorg, Foss argues the popular 'AI flattens organizations' claim is theoretically premature: AI cuts transaction costs on both sides of Coase's make-or-buy boundary simultaneously (cheaper market search/contracting favors smaller firms; cheaper internal information synthesis favors larger ones), so the net effect on firm size is genuinely ambiguous, not predetermined. Using Garicano's knowledge-hierarchy model, he argues AI unbundles what hierarchy has always bundled together — information routing, expertise access, coordination, monitoring, judgment, and authority — so these can now move independently: a firm can flatten informationally while staying hierarchical in ownership, or decentralize operations while centralizing strategic control. Applying Hayek, he notes the same visibility infrastructure that enables genuine decentralization could equally enable a Taylorist panopticon of centralized control — the technology doesn't determine which. Drawing on Knight, he argues that as AI commoditizes analysis, coding, search, and forecasting, what stays scarce is entrepreneurial judgment — deciding which opportunities are real and which resources to commit under genuine uncertainty no data can resolve — which is why ownership and strategic authority should concentrate even as operational autonomy disperses. His practical illustration: a German customer-retention scenario where AI assembles temporary problem-based teams that dissolve once resolved. Bottom line: the truly AI-native firm — reorganized from first principles around what AI actually makes cheap versus what stays scarce — 'perhaps still doesn't exist' yet.
Editor's opinion
The real contribution is refusing the easy 'AI flattens hierarchy' story with an actual mechanism: AI cuts costs on both sides of Coase's boundary at once, so the net effect on firm size is genuinely ambiguous, not a foregone conclusion — a real correction to a claim getting repeated uncritically across a lot of AI-and-organizations commentary right now. The panopticon point is the sharpest part: the same visibility infrastructure that enables genuine delegation can equally enable centralized control Frederick Taylor could only have dreamed of, and nothing about the technology itself decides which way an organization goes — that's a judgment call for leadership, which loops back nicely to his own point about entrepreneurial judgment being the thing AI can't commoditize.
Key concepts
Coase's theory of the firm · knowledge hierarchies (Garicano) · Knightian uncertainty · judgment-based theory of entrepreneurship · unbundling of hierarchy