What Happens When Everyone Can Build Software With AI
AI coding agents are letting non-engineers ship real apps. Here's how the shift is playing out, based on YC's Lightcone and Supabase's own numbers.
What’s actually changing?
AI coding agents have gotten good enough that people with no software engineering background are shipping working apps, not toy demos but things with real databases, real integrations, and real users. The clearest evidence comes from infrastructure providers watching their own growth curves bend. Supabase, the open source Postgres backend used by tools like Claude Code, says it has grown to 13 million signups and is spinning up over a million databases a week. A year earlier it had around 3 million signups and roughly 8 million databases under management. That kind of jump isn’t just more developers showing up. It’s a new category of user: the “builder” who isn’t a software engineer but can now describe what they want and get a working product.
TL;DR
- Supabase’s growth numbers tell the story: signups went from about 3 million to 13 million in a year, and databases under management grew from 8 million to nearly 50 million, with revenue on track to grow 4x.
- A new user segment is driving it: people without coding backgrounds, described on the Lightcone podcast as “builders,” are now able to go from idea to working software without convincing an engineer to help them.
- Domain expertise matters more than code now: the hard part was never typing syntax, it was knowing what to build, and that knowledge now belongs to people who previously had no way to act on it.
- Open source infrastructure has an edge with agents: Supabase’s team argues that agents prefer tools they can inspect when something breaks, and that enterprises will increasingly demand inspectable source code for security reasons.
- Personal software is becoming a real category: founders and non-engineers are building single-user tools for finance, health, and inbox triage that rival dedicated consumer apps, because the software only has to work for one person.
- The ceiling keeps moving: investors expected product-led growth to taper off as it ran out of technical users to convert, but Supabase says that pattern broke down this year as non-technical builders kept arriving.
- Name
- Maria Alvarez
- Date of birth
- SSN
- Diagnosis
- Balance due
- $412.80
Who counts as a “builder” now?
The term “builder” is doing a lot of work in this conversation. On the Lightcone episode, Supabase CEO Paul Copplestone draws a line between traditional software engineers, who still make up a large share of activity, and a fast-growing group of people who have ideas and domain knowledge but never had the technical ability to execute them. These are people who open an AI coding tool, describe what they want, and get something that works well enough to use or even sell.
One example raised in the discussion: a taxi driver in Brazil built his own ride-hailing app for his local market, from his phone, without any coding background. Another: a fintech founder who exited his company for a large sum used his domain knowledge of banking integrations to build personal finance software for himself, the kind of tool that previously required a dedicated company and team to build.
The common thread is that these builders already knew the problem cold. What they lacked was the ability to translate that knowledge into software. AI coding agents removed that bottleneck.
Why is this happening now and not two years ago?
The honest answer is that models got good enough to build without constant hand-holding. Supabase’s team points to a shift in how it structures its own product to meet this moment: the company describes three phases of its existence, starting as a dashboard product, moving to an API-first company, and now operating as what it calls a “code-first” company, built around the command-line interfaces and workflows that AI agents actually use.
Being chosen as a default database inside popular AI coding tools mattered too. Supabase’s team credits timing: becoming a default option in large language model tool integrations meant that when someone asked an agent to spin up a backend, Supabase was often the path of least resistance.
Does open source give AI agents an advantage?
Supabase is fully open source, and its team argues this wasn’t incidental to its growth with agents, it was central to it. The reasoning given is practical: when an agent hits an error it can’t resolve, having access to the actual source code lets it debug instead of hitting a wall it can’t see past. Closed systems don’t offer that option.
There’s also a forward-looking argument about enterprise adoption. As security scrutiny around AI-built software increases, the expectation is that companies will want to inspect the source code of the infrastructure their agents depend on before trusting it in production. Closed source vendors may find that a harder sell going forward.
What are people actually building for themselves?
Some of the most interesting examples from this trend aren’t startups at all. They’re personal software: single-user tools built by individuals purely for their own lives, with no intention of selling access to anyone else.
Without shadow IT.
Copplestone describes building what he calls a personal “control tower,” a set of tools covering email triage, finance tracking, and health monitoring, all backed by a single personal database. He argues the finance section alone, built for one person, reaches feature parity with consumer products like Mint or Monarch, and in some ways surpasses them because it has personal context a generic app never could. It can look up who sent a payment by cross-referencing a personal notes system and categorize the transaction correctly without being told how.
This points to something structural: software built for an audience of one used to make no economic sense, because the cost of building it exceeded any benefit beyond generic tools. AI coding agents collapse that cost close to zero, which makes “built only for me” a viable category for the first time. The data behind that software, stored in something like a personal database, becomes the valuable, durable asset. The individual apps and agents layered on top become replaceable.
Is there a ceiling to this growth?
Supabase’s team says they expected one and haven’t found it. Early investors reportedly warned that product-led growth strategies tend to plateau once they exhaust the pool of technically sophisticated users, forcing a pivot toward sales-driven enterprise motions. That pattern is said to have held roughly a year or two ago. This year, according to the company, it broke: signups, activation, and conversion are all climbing together, which isn’t the shape a maturing, saturating market usually takes. The explanation offered is that the pool of potential builders isn’t shrinking, it’s expanding, because the skill required to participate dropped sharply while the number of people with unmet software needs did not.
Frequently Asked Questions
What is a “builder” in this context?
It refers to someone without a traditional software engineering background who uses AI coding agents to design, build, and ship functional software, often solving a problem from their own domain expertise rather than a general technical skill set.
Do AI coding agents replace software engineers?
Based on this trend, engineers still represent a large and active share of usage on platforms like Supabase. The change is additive: a new population of non-engineer builders is growing faster than the engineer segment, not replacing it outright.
Why does open source infrastructure matter for AI agents specifically?
Agents benefit from being able to inspect source code when they hit errors they can’t resolve through documentation alone. Open source tools let them debug directly, which closed source tools don’t allow.
What kinds of apps are non-technical builders actually making?
Examples mentioned include a locally built ride-hailing app made by a taxi driver and personal finance and health tracking tools built by individuals for their own single-person use, sometimes rivaling dedicated commercial products in functionality.
Is this trend specific to one tool or platform?
The trend is broader than any single product. It reflects a shift in AI coding agents generally becoming capable enough to let non-engineers build working software, with infrastructure providers like Supabase serving as one visible data point for how fast that shift is moving.
