When a developer tool starts drawing more casual, consumer-style signups, its numbers usually get worse. Fewer of the new people set the product up, and fewer of them pay. Ant Wilson, co-founder of the database company Supabase, said that is the usual pattern. "But both of those things went up," he told swyx on the Latent Space podcast, in an episode published October 28, 2026, where he appeared with fellow co-founder Paul Copplestone.
Supabase sells a backend: the database and related services that sit behind an app and store its users, records and files. It is built on Postgres, a widely used open-source database. More and more of Supabase's new customers now arrive through AI agents, programs that write and set up software for someone who describes what they want. People call that way of building "vibe coding."
Why the funnel held up
Software companies track a signup funnel. People first sign up. Some then activate, meaning they actually start using the product, and a smaller share later convert to paying customers. Wilson said Supabase's funnel metrics showed both later steps improving as AI-driven signups grew.
He gave a simple reason for activation. A human developer might create an account and never come back. When an agent builds the app, "the agent does it for you," Wilson said, "like in one step." Signup and setup happen together, so activation is "almost a hundred percent" and "friction's gone." Those users then converted to paid plans "at a higher rate than our traditional developer user," he said. He gave no conversion percentages. He added that he expected other developer tools were seeing the same thing.
The scale behind those signups comes from the company's own figures. In an October 2, 2026 funding announcement, Supabase reported more than one million new users and four million new databases a month, and said agents or AI tools created 70% of the new databases.
Money in the first week
Asked how quickly people now scale what they start, Wilson pointed first to how quickly they start making money. In Supabase's early days, he said, you chose your database, built your product, and added billing "at some point." Now agents can wire in Stripe, the payments service, while they build the app, and Supabase sees customers start a project and begin "generating serious cash" within the same week.
That speed changed how the founders saw vibe coding. At first it wasn't obvious to them that the trend would last or change how software gets built, Wilson said. Then they saw a builder make what he recalled as $10,000 within the first few days of launching a product. "We were like, wow, this is, this is real," he said. "People are making real things that generate real money."
Treated differently from NFTs
Supabase had ridden earlier waves, and Wilson used one of them as the contrast. When NFTs were huge, the company added sign-in with Ethereum and sign-in with Solana, but it didn't put much developer time behind them "because everyone kind of knew it was going to go away at some point."
Vibe coding got a longer debate inside the team about how far to lean in. Wilson said the timing was awkward. Supabase's biggest customers were stretching the limits of what it offered, so the company already needed to move toward larger customers. It decided to do both: keep building for small prototypes and for very large deployments, and hire the extra people that required.
Three chapters of AI
In the episode, Supabase's experience with AI was laid out in three chapters.
The first was embeddings. Embeddings are lists of numbers that capture the meaning of text or images, so a system can find similar items. As specialized "vector databases" for storing them appeared, Supabase argued that Postgres could do the job, so developers wouldn't need a second database. The conversation credited an extension by Andrew Kane, an independent developer. That is most likely pgvector, which adds vector columns and similarity search directly to Postgres. The episode described that period as a good growth driver for Supabase and for the Postgres ecosystem. Embedding storage has kept changing since then. The discussion named turbopuffer as one example, a service that keeps its data in object storage with a fast cache in front.
The second chapter was the vibe-coding wave at the end of 2024, when app builders such as Lovable and Bolt arrived. The open questions were whether Supabase wanted to serve those platforms at all, and whether it should build its own vibe-coding product on top of Supabase. In the episode, the decision was tied to the company's mission of storing the world's data, and to being there when people get started so they never want to leave. The early projects were mostly toys, but they fit that mission, and the bet was that the tools would get better.
The third chapter started in 2026, when the models improved. Supabase's signup rate began rising quickly. A jump like that usually comes with heavy marketing spending, but the episode said Supabase spends almost nothing on marketing beyond memes and similar material.
Enterprise comes fourth
Asked about the company's top objectives for the year ahead, the answer came as a ranked list: win the strategic labs, win the models, continue the product-led growth push, and then enterprise in fourth place. Product-led growth, or PLG, means customers find, try and buy a product themselves rather than through a sales team.
Wilson explained why PLG still ranks so high. When the company started, the founders thought PLG alone could take them to "100 million," maybe 200 million, before they would need a different way of selling. He did not say what the figure measured. "Because of AI, like that ceiling just lifted and we don't know where it got lifted to," he said. "Is it a billion? Is it two billion?" Supabase still wants to pursue large customers too. "Maybe we're a bit greedy," he said.
Eighteen months before the episode, the company was already growing fast, Wilson said. Had someone asked him then what it could do to grow its signups tenfold, "I would have said there's nothing."
The episode also described this year's revenue. At a company offsite at the start of 2026, Supabase set an extremely ambitious revenue target, and someone warned that the team would be disappointed when it missed. It took about four months to reach. No figure was given. The company's tagline has changed along the way. It used to promise you could build in a weekend and scale to millions. Now it is "prototype to petabyte."