AI agents now drive 70% of new database growth, not human engineers
People across industries no longer need to convince engineers to get started: AI lets people who understand business build complex systems directly, and there's currently no ceiling on the growth of this new builder cohort.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
In one year, managed databases grew from 8 million to 50 million
Superbase registered users rose from roughly 3 million a year ago to 13 million; over 1 million new databases are created weekly; total managed databases approach 50 million (up from 8 million a year ago); revenue is projected to grow 4x this year. The driver isn't just that AI helps software engineers write code faster—the real inflection point is that people without engineering backgrounds can now build directly on Superbase for the first time.
— PaulUnusual growth curve: conversion metrics improved during explosive user scaling
Usually when user counts surge, activation and conversion metrics dilute and decline—it's an almost-universal law in product growth. But this year all of Superbase's related metrics rose simultaneously, because models became strong enough that people who couldn't write code could now build it. The old investor claim that product-led growth tops out at some revenue threshold and must shift to sales-driven growth has not held true as of this year.
— PaulScaling to 1.2 billion monthly database creations is uncharted territory for the industry
Paul did arithmetic using ChatGPT's 1.2 billion weekly active users: if each user wanted to build something once a month, that's 1.2 billion databases/month—a scale the database industry has never had to handle. The company acquired Terso (compatible with Postgres and SQLite), launched lightweight Superbase Spark, positioned as ‘from prototype to petabytes’, and maintains both a multi-tenant architecture that scales to petabytes and an architecture for lightweight starter scenarios.
— PaulCode at home: engineers were convinced, not mandated
The company has no AI usage policy—employees use whatever tools they want. The result: 10% of engineers writing 80% of code using AI grew to 60% of engineers doing the same within six months—not through force, but because engineers were gradually convinced by the model's actual output quality. As AI-generated code increased, the team also massively increased the test volume they wrote to verify correctness—the same pattern shows up at other YC companies like Blacksmith and UbiCloud.
— PaulThree transitions shaped the product: dashboard, API, then code-first
Each transition had a concrete trigger. The first came when software engineers themselves found databases too hard to use, so they built a dashboard (like Airtable) on top of Postgres. The second came in 2021 when a customer's bulk-database request forced out an API, just as AI site-builders like Bolt and Lovable exploded in 2024. The third happened this year: the team noticed almost no one was accessing the dashboard anymore, so they rebuilt the CLI and entire product as ‘infrastructure as code’, naturally matching how agents work.
— PaulWinning agents takes two phases: selection, then preventing switching
Superbase rewrote error messages designed for humans—the ‘HTTP 500’ style—to tell agents directly how to fix the problem and which docs to read. Winning the AI-tools business has two steps: first, get selected as the default database, then when an agent gets stuck (and all agents eventually get stuck), make sure it doesn't switch to a different tool. Give it self-healing capabilities. Rewriting product details specifically for agents rather than humans now takes up significant engineering effort.
— PaulPersonal knowledge bases need retrieval structure, not just a bigger context window
GBrain's builders found that no context window is large enough—100k tokens ≈ three Harry Potter books, but getting an agent to pull exactly the right three books when answering a question is hard. So they used Postgres's pgvector to build a retrieval system, organizing unstructured data into a graph structure directly in the database, then ran automated roles like Cappy 24/7 in eval loops, continuously testing and improving retrieval accuracy—essentially a very primitive recursive self-improvement loop for one specific task.
Vertical SaaS fundraising is hard; small teams hit high margins with AI
The market has an anomaly: vertical SaaS companies with $5–10 million annual revenue struggle to raise Series A (this was a stable threshold nine to twelve months ago), but horizontal and AI-related projects sail through. Meanwhile, a 15-person team can hit $25–50 million in annual revenue, $2 million in ARR per person, and 50% margins—no longer needing a huge funding round to hire bodies. ‘Didn't raise funding’ went from a source of embarrassment to something worth envying.
In their own words · checked verbatim
I think actually if databases under management now are approaching 50 million, and this time last year it was 8 million.
Paul1:05
usually when this happens, you see a lot of the other numbers start to decline, activation, conversion. But actually this year, all of them are kind of going up and to the right, which is a bit insane just because the models are getting so good at building.
Paul4:14
Well, one kind of cool example was at our conference. We had the conference here and afterwards, this guy from Brazil came up to me and he's built an entire taxi app from his phone. And it's kind of like a little mini Uber for his local market.
Paul8:24
because there's like two stages to winning the agents the first is being picked by the model and then the second is like if they get stuck you don't want them to switch away
Paul30:57
a million tokens is actually an incredible working memory like most human beings only have like seven plus or minus two things that they can keep in their head whereas like a million tokens is like basically uh like three harry potter books
It was like if you've got 15 people, 2 million ARR per head and with 50% margins, you're all making a million a year. I mean, that's an amazing outcome for a lot of people.
Databases have been around for 50 years. I think they'll be around for another 50 years.
Paul45:40
Figures
| Superbase registered users | 13 million (3 million a year ago) | 1:05 |
| Total managed databases | close to 50 million (8 million a year ago) | 1:05 |
| Projected revenue growth this year | 4x | 1:05 |
| ChatGPT weekly active users | 1.2 billion | 14:34 |
| Engineers using AI for code, six-month shift | from 10% of engineers writing 80% of code to 60% of engineers | 17:36 |
| AWS revenue from startups | about 40% | 24:48 |
| 15-person team annual metrics | $2 million ARR per person, 50% profit margin | 43:32 |
Glossary
- Recursive self-improvement
- AI refines itself through automated loops of testing and iterating; here, continuous eval runs to optimize retrieval performance.
- Vibe coding
- Speaking requirements in conversational style and having AI directly generate code without scrutinizing implementation details.
- Lindy effect
- The longer something has existed, the longer it's expected to continue existing.
- PLG (product-led growth)
- Growth driven by the product's inherent appeal and conversion ability, not by a sales team.
How to listen
Founders, product managers, and early investors tracking AI-native infrastructure, developer tools growth curves, and the trend of everyone becoming builders.
The intro ([0:00]) is a mashup of full-episode clips; the content repeats later, and you can skip straight to [1:05].