Interview, Fireside Chat
Why Everyone Is About To Have Their Own Database
- Predicts a future where global AI access enables all individuals to act as builders, customizing apps and UIs without external dependency, potentially leading to 1.2 billion monthly databases and a market with "no ceiling" for growth.
- Forecasts "huge" revenue growth over the coming year driven by a shift from 8 million to approximately 50 million databases annually, with "70% of new activity" attributed to AI agents.
- Projects that by 2027, assuming no major disruptions, the company will experience a significant growth trajectory, with internal AI coding adoption reaching 60% of code written by agents within a six-month span.
- Anticipates that in one year, software development will be highly automated, allowing users to build complex sites like "open usd.dev" in 20 seconds using minimal prompting.
- Plans to launch "Superbase Spark," a SQLite-compatible tier for prototyping that scales to petabytes, complementing a core architecture designed to handle petabyte-scale loads for enterprises and personal agents.
- Shifts strategy toward a "code first" model with re-architected CLIs and sandbox environments to meet the demands of AI agents that primarily interact via code rather than GUIs.
- Identifies a significant risk that closed-source code will face adoption barriers among enterprises requiring source inspection for security, reinforcing the critical role of open source for debugging and trust.
- Notes a challenging fundraising environment for vertical SaaS founders at the $5-10 million ARR stage, contrasting this with opportunities for horizontal/AI investments and profitable companies with 10-15 people reaching $50 million ARR.
- Envisions a "prosumer" market explosion where individuals own personal databases for health, finance, and relationships, serving as a durable "store of record" that avoids the need for aggressive pivots to enterprise.
- Outlines architectural challenges in managing billions of new databases, including the need for dynamic schemas, automatic index creation, and self-healing error messages to prevent agents from getting stuck.
- Describes an internal evolution where "evals," "auto-research," and "captains" automate improvement loops, shipping code to master by default based on statistical improvements while logging failures in markdown files.
- Predicts that "dynamic retrieval" and vector search capabilities (e.g., "G-Brain" utilizing PostgreSQL and pgVector for 30-40 gigabytes) will allow agents to autonomously manage data structures, graphs, and context.
- Highlights the potential for blue-collar workers and non-engineers to become entrepreneurs by using AI to build sophisticated personal applications like tax, finance, or health tracking apps without hiring engineering teams.
- Emphasizes a business model focused on "getting customers on day one" to scale rapidly, leveraging partnerships and "dogfooding" to identify gaps before customers do.
- Assumes databases will remain a durable category for the next 50 years, advising founders to invest in categories that are resilient to AI era shifts.
- Acknowledges a potential bottleneck where AI model capabilities grow parabolically while the number of humans capable of guiding societal impact remains constant.
- Recognizes that negotiating better terms with cloud providers like AWS could enable scaling to 1 billion new databases per week, though there is a prevailing desire to build proprietary engines.
- Observes that builder activation conversion is currently rising across the board, indicating the "ceiling" on builder growth is not yet reached, unlike previous growth surges where conversion dipped.