Korva is a scroll of the internet’s best ideas.
Database migration became trivial
Why databases stuck
Databases were among the stickiest businesses in software. App developers built against the specific interface of one database, and as the data piled up, moving an app to a different database became a giant, painful project. That stickiness paid for Oracle, SQL Server, and a long list of smaller players that earned great margins.
What agents changed
Now Claude or Codex does the building against those interfaces, and database interfaces are very well specified. AI is very good at well-specified things, and agents never tire of the monotonous work of translating one specification into another. Moving a database used to be the number one thing nobody would do in software. With money put against it, it is now close to trivial.
What wins now
Nobody has to build their own database, and databases still matter. What changed is the list of qualities that make one great. Experiments cost far less, so far more applications get started, and a database has to scale from almost zero usage to enormous if the app takes off. Teams spin these things up, spin them down, and tear them apart over and over, so cost matters much more. Cost, scaling from zero to infinity, and portability now decide who wins, where a license procured for one product on one kind of hardware once did.
Why were databases so hard to leave?
SQL is a standard, but every vendor extends it with its own features, such as Oracle's PL/SQL and Microsoft's T-SQL. Applications pile up stored procedures, triggers, and query habits written in those dialects. A move also means copying every row and proving that each query still returns the same answer. Large migrations away from Oracle have commonly been multiyear projects.
How can a database scale from zero?
Serverless designs split storage from compute, so the compute can shut down when nobody is using the app. Amazon Aurora Serverless is one example of a database sold this way. Billing follows usage, which means an idle app costs close to nothing. Traditional servers were provisioned for peak load and billed around the clock.
Why are database interfaces so well specified?
SQL was standardized by ANSI in 1986 and by ISO the following year. PostgreSQL and MySQL publish their wire protocols in public documentation. Those written rules give a model an exact target to translate from and to. The SQL standard has since been revised repeatedly, with major editions in 1992, 1999, and 2003.
More from Invest Like the Best with Patrick O'Shaughnessy
