In an episode of the Lightcone podcast from Y Combinator, the hosts sit down with Paul Copplestone, co-founder and CEO of Supabase, the open-source Postgres backend that has become a default database inside coding agents. Supabase started in the YC summer 2020 batch and, by his account, now has about 13 million sign-ups and nearly 50 million databases under management. The conversation covers agents as customers, a new wave of "builders" who are not software engineers, personal databases, scaling to billions, and what he has learned about building products that agents want. The video runs about 49 minutes.
This article is our own summary, with short quotes and timestamps from the video's transcript. Watch the full interview above for everything that was said.
The short version
- Agents are now the main users. A host says roughly 70% of new activity is from agents, and Copplestone calls the shift "wild" (0:00 to 1:35).
- Growth numbers. About 13 million sign-ups versus around 3 million a year earlier, revenue growing about 4x this year, roughly a million databases created per week, and database count heading for 50 million from 8 million (1:35).
- Code-first wins. Supabase went from dashboard company to API company to code-first company because agents prefer the command line (3:03, 21:03).
- Builders, not just developers. People without engineering backgrounds are building real software, and conversion and activation are rising, not falling (4:31 to 9:05).
- Everyone may have a personal database. His own "control tower" holds email triage, finances, health and files (10:33 to 13:31).
- Scale from prototype to petabytes. A lighter tier for tiny workloads and a scale-up path for large ones (15:00 to 16:32).
- Make something agents want. Pick the right category, fix error messages for agents, and test with evals (28:36 to 33:00).
What he said, by topic
Make something agents want (0:00 to 3:03)
A host riffs on the YC slogan "make something people want" and says it is increasingly "make something agents want." Copplestone says the inflection toward agents this year is huge. Supabase now sees a million databases created a week. He describes three chapters: first a dashboard company, then an API company, and now "code first," offering the modes agents want, mostly the command line. Being chosen as a default inside large language models was also a big boost.
Why open source helps (3:03 to 4:31)
He says he can't know the counterfactual but believes being open source was critical. Agents, the hosts suggest, like open source because when something breaks they can read the source. He adds that enterprises will increasingly want to inspect code for security, and may avoid closed-source vendors.
The rise of builders (4:31 to 9:05)
The surprise this year, he says, is that sign-ups are accelerating while activation and conversion also go up, which is unusual. The reason: models are good enough that people without software backgrounds can build. The hosts describe a fintech founder who built personal finance tooling for himself, and domain experts who can finally build for their niches. One example Copplestone shares is a taxi driver in Brazil who built a small ride-hailing app for his local market from his phone.
Personal software and personal databases (9:05 to 15:00)
He describes a "control tower" for his own life: email triage, spending, health and workouts, a CRM and file uploads, all stored in Supabase. He argues the database is the valuable record and that you should own it and attach agents on top, rather than boxing your data into one agent. He says he would not have wanted a personal database a year ago, and now it feels like the most valuable thing. He thinks everyone with access to AI will become a builder of some sort.
Scaling to billions (15:00 to 18:02)
Copplestone says the mission has not changed but the aperture has widened. If a billion people each launched something a month, that would mean over a billion databases a month, a scale the database world has never faced. Supabase acquired a company whose architecture is compatible with Postgres and SQLite to offer a lighter tier for getting started, with a path "from prototype to petabytes." Serving both ends at once is, he says, difficult but fun.
AI inside Supabase (18:02 to 19:58)
A host cites a poll where the share of engineers writing most of their code with agents rose from 10% to 60% in about six months. Copplestone says Supabase has no AI policy and lets people use whatever tools they want, and that heavy testing has become important to trust the result.
Lessons from three eras (19:58 to 27:29)
He recalls that a database company looked contrarian when they joined YC, and that the first dashboard looked like Airtable on top of Postgres. The API shift came from a customer who wanted to launch many databases, which turned out to be right when app builders arrived in 2024. This year "no one visits dashboards anymore," so he asked the team to go code-first. He credits earlier backend platforms such as Parse and Firebase for blazing the trail. On enterprise versus early-stage, his model is the Stripe-style approach: be there on day one and scale with the customer so they never want to leave.
Building for agents and dogfooding (27:29 to 33:00)
He says he is bullish on personal agents and tests his own product by talking to an agent and refusing to look at the code, to behave like a non-technical user. A key tactic is making error messages tell agents how to fix the problem, because there are two stages to winning agents: being picked by the model, then not being dropped when they get stuck. Supabase runs evals across models, and more of its engineering effort goes into testing use cases than building.
Agent-first design and evals (33:00 to 38:42)
A host explains how GBrain, an open-source retrieval system for personal agents, runs on Supabase, using Postgres with vector search and a graph-like structure. The host also describes an eval set and a loop where agents try improvements and ship the ones that raise scores, calling it a primitive form of recursive self-improvement for one use case.
The future of coding (38:42 to 42:36)
Copplestone describes talking to his personal agent by voice to build a developer site, refreshing the page and finding the structure improved. He tries to "code like a normie" so he knows how Supabase will work for non-experts next year. The host sees a gap between model capability and the number of people who know what to build.
More entrepreneurship (42:36 to the end)
He is "pretty pumped about 2027," expecting more entrepreneurship and fulfilling lifestyles. The hosts note that vertical SaaS founders are struggling to raise a Series A while small teams can reach large revenue. His advice: choose durable categories (databases have lasted 50 years), keep one eye on the future, stay adaptable, dogfood your own tools and leave capacity for partner ideas, knowing many bets will flop.
Chapter guide
- 00:00 Intro
- 01:18 Growth and 4x Revenue
- 02:35 Three Eras of Supabase
- 04:34 "Builders" vs. Traditional Devs
- 08:42 Personalized Apps
- 12:12 Personal Databases & Data Ownership
- 15:56 Scaling to Billions
- 17:58 AI Adoption At Supabase
- 19:58 Lessons Learned Along The Way
- 27:29 Building For Personal Agents, Dogfooding
- 30:01 Agent-First Design and Evals
- 38:42 The Future Of Coding
- 42:36 Opening Up More Entrepreneurship + Outro
What it means for your business
- Assume agents are customers. Make your product work from the command line and an API, not only a dashboard.
- Write error messages for agents. Tell them how to fix the problem or where to read the docs.
- Test with evals. Measure how different models use your product and fix the gaps.
- Serve non-engineers. Domain experts with pent-up ideas are a fast-growing audience.
- Own your data. A personal or company database you control lets you swap agents without losing context.
- Pick durable categories. Build where demand will outlast any one technology wave.
Video details
- Title: What Happens When Everyone Can Build Software
- Guest: Paul Copplestone, co-founder and CEO of Supabase
- Published by: Y Combinator on YouTube (Lightcone), October 30, 2026
- Length: about 49 minutes
- Watch: youtube.com/watch?v=vl7HiMdGfjo
- Note: quotes come from the video's automatic transcript and are lightly cleaned. Check the video for exact wording.
Want to put this to work?
We help owners turn AI trends into a plan, a product and a launch. Book a free call or see our services.