Basic idea of double entry bookkeeping is that if you borrow money, it represents both an asset to you which you can spend, but also a liability (for the lender the reverse is true) because you have an obligation to repay it later, and that any accounting system should reflect all this lending that goes on. Knowing what they owe and are owed in future is as important to banks as knowing what payments they've just received and just made.

She's contrasting that with a "techie" view where all that matters is securely and immutably recording the fact that a borrower has received the "coin" (which represents mainstream blockchain implementations, though I'm sure blockchains could be designed to reflect debt relationships and even enforce debt repayment), and suggesting that what the distributed consensus actually records is only a subset of the relationships a centralised accounting/clearing system deals with.

She also pointed out that our financial system being built around debt relationships actually guarantees there will be some non-zero level of demand for dollars/Euros/Yen etc in future, whereas building the Bitcoin ecosystem around who holds the Bitcoin at a given point in time doesn't actually guarantee people will still want them tomorrow (especially if they owe repayments on dollar-denominated debt to buy BTC or mining equipment)