Y CombinatorWhy Everyone Is About To Have Their Own Database
Y Combinator
Agents Drive Growth 0:00
Paul Coppstone says Superbase has moved into a new phase as agents take over more of the work. He says about 70% of new activity now comes from agents, and that signups, revenue, and databases under management are all rising fast. Last year, there were about 3 million signups; now there are 13 million. Databases under management have climbed from 8 million to nearly 50 million, and new databases are being created at a rate of about a million a week.
Builders Without Barriers 2:30
The change is not only about software engineers. Coppstone says more people with ideas are coming online, including domain experts and non-engineers who can now build with tools like Cloud Code. Superbase has shifted from a dashboard company to an API company and now to a code-first company, with CLI use and open source support helping agents work well. He gives examples of a former fintech founder building personal finance tools, and a taxi driver in Brazil making a mini Uber for his local market.
Scale for new builders 15:00
The mission stays the same, but the customer base has widened a lot. The company says it now launches about a million databases each week, yet it is planning for a world where millions of people may want to start something at once. That means building for a scale that database systems have not had to face before. The answer is to keep the same simple start and be ready to grow far beyond it.
Two ends of the market 16:01
To cover that range, the company is splitting its product line. One side is for small, getting-started work, with a PostgreSQL and SQLite compatible path under the name Superbase Spark. The other side goes the opposite way and scales up to petabytes. That lets people begin with one experience and keep using the same system as their business grows.
Agent-friendly errors 30:30
The focus is on fixing error messages so agents can recover on their own. If an agent gets stuck, it should not have to switch away. It should see the error, understand what went wrong, and get a clear hint about the next step. That means reworking tools and messages that were once made for humans and developers.
Gbrain and retrieval 35:01
Gbrain is described as a company brain built from retrieval and graph structure. It uses hosted Postgres, pgvector, and SQL graphs to handle large corpora fast, including 50, 100, or 300 gigabytes of data. An agent asks what it needs, the system traverses the graph and vector search, and it returns a report to load into context. The same setup is used to test browser use and self-manage through the database and APIs.
Stay adaptable 46:01
He says the team cannot assume the answers stay fixed because the market changes so fast. The goal is to keep adapting faster than others, stay close to how developers are building now, and keep capacity to shift when the next wave arrives. He points to founder-led focus at Stripe as the model.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
