Alina Schanz

Rollups now pay Ethereum very little to publish their data. A handful of builders produce most blocks. What users pay is set less by the base layer and more by the operators in between.

That is where I expect durable infrastructure businesses: teams that control ordering, routing or distribution, not raw capacity.

Most agents today run on a borrowed human account with all of its authority. What they need is duller and harder to build: keys that can spend without owning, budgets that expire, and a record that someone else can audit afterwards.

The same holds for machines in the physical world. A robot that acts on its own still needs an operator who can see what it did and take back what it was allowed to do.

Open models and distributed training let people who have never met train or serve a model together. Whether that works comes down to accounting: who contributed which compute, how anyone checks it, and who gets paid. That is where crypto has something real to offer AI.

With any compute network, my first question is what happens when a claim of work is false: who notices, and who pays.

Privacy systems have become good at hiding transfers and still leak much of what surrounds them: who handed a transaction to the network, from where and when. The harder design question is who has to be identified, and to whom.

An identity product gets my attention when it can say plainly what it reveals, to whom, and who can be made to disclose it.

On the same day, the same stablecoin dollar can earn rates many times apart in neighboring lending markets, and some of the largest onchain transfers are loans that leave and return within a single block. Headline volume rarely says which is which.

I look for mechanisms that still work after the subsidy ends.