A conversation with Andrey Fradkin, a professor who studies the economics of digital markets and AI, and is the co-host of the Justified Posteriors Podcast.
The trust infrastructure point is the one that keeps me up at night (honestly). When effort-based signaling breaks - when anyone can generate polished content, polished bids, polished credentials - reputation systems need to do a lot more work.
The transaction cost framing is useful though. I hadn't thought of it that way: agents reduce search cost but they also reduce friction, and friction was doing invisible trust work. A human had to show up, answer questions, prove availability. An agent just... responds instantly, always, forever. What does credibility even look like when the cost of appearing credible goes to zero?
Exactly. My guess is credibility starts shifting away from surface-level effort and toward things like reputation, verified identity, historical performance, and how well an agent is actually calibrated to the user it represents.
Agents are distinct by their capture of scarce resources—context or actions.
Any agent distinct by context is an information monopolist, who maximizes profits by vertical integration, not eroding their scarcity by telling everyone their alpha.
The coasean effect is actually the opposite. Firms with high value context *become very large*.
I mean, it’s trivially true that traditional transaction costs fall.
But negative externalities from transacting are clearly also transaction costs.
I think a helpful way to imagine it is that every agent or company is an instance of Claude with identical capabilities. If they’re profit-maximizing, why would they transact if transacting is informative and weakens an advantage?
Illustrative: notice more than half of all public market trade volume happens off-market / in dark pools.
The trust infrastructure point is the one that keeps me up at night (honestly). When effort-based signaling breaks - when anyone can generate polished content, polished bids, polished credentials - reputation systems need to do a lot more work.
The transaction cost framing is useful though. I hadn't thought of it that way: agents reduce search cost but they also reduce friction, and friction was doing invisible trust work. A human had to show up, answer questions, prove availability. An agent just... responds instantly, always, forever. What does credibility even look like when the cost of appearing credible goes to zero?
Exactly. My guess is credibility starts shifting away from surface-level effort and toward things like reputation, verified identity, historical performance, and how well an agent is actually calibrated to the user it represents.
AI *increases* transaction costs.
Agents are distinct by their capture of scarce resources—context or actions.
Any agent distinct by context is an information monopolist, who maximizes profits by vertical integration, not eroding their scarcity by telling everyone their alpha.
The coasean effect is actually the opposite. Firms with high value context *become very large*.
I mean, it’s trivially true that traditional transaction costs fall.
But negative externalities from transacting are clearly also transaction costs.
I think a helpful way to imagine it is that every agent or company is an instance of Claude with identical capabilities. If they’re profit-maximizing, why would they transact if transacting is informative and weakens an advantage?
Illustrative: notice more than half of all public market trade volume happens off-market / in dark pools.