Fabian Hedin co-founded Lovable, the platform that turns plain English into working software. At his session at Anthropic's Code with Claude conference in London, he put the scale in numbers. The sites people build on Lovable pull 600 million visits a month, more than Lovable's own site gets. "People are building things that kind of out-succeed Lovable combined," he said.

Lovable was built for people who never learned to program. "We called that the 99% right there," Hedin said. The mission started with a tool his co-founder Anton posted on GitHub called GPT-Engineer, which became the fastest growing repository on the site. Today the platform hosts 50 million products and adds 200,000 new ones every day. Lovable is the loudest name in vibe coding, but Replit, Base44, and Bolt are chasing the same non-technical users.

How vibe coding works at 600 million monthly visits: lessons from the Code with Claude conference

Non-coders are shipping production software

The people who are vibe coding rarely fit the developer stereotype. "You have kids working on it... and we have people up in the Fortune 500 using this tool as well," Hedin said. When the team segmented users by job, the largest group called themselves engineers. Even the people who can code would rather work this way.

Vibe coding doesn't mean you have to compromise. "We really want to build production grade software and really chase the frontier of what's possible," Hedin said. The team could have settled for quick prototypes. Instead they build full applications for people who never learned to code. That combination is what makes Lovable’s engineering hard.

The last 10% is where vibe coding breaks

Hedin shared a programmer's joke that has outlived the era it came from. "The last 10 percent of code takes 90 percent of the time," he said, then the punchline: "the first 90 percent also takes 90 percent of the time. So you end up with 180 percent of the time." The joke makes a serious point. A project's first draft comes fast, and finishing it always takes far longer. Vibe coding works the same way. Finishing is the wall.

The wall hurts non-coders most. A developer who hits a bug can open the code and dig in. Someone building on the abstraction cannot. Lovable measures this with a flag it calls is_stuck. "It will be true if you're asking for the same thing three times in a row. So if you're asking, fix it, fix it, fix it, we will assume you're stuck," Hedin said. Getting stuck is the worst thing that can happen to a non-technical builder.

Every app makes the next one better

Lovable built continuous improvement into their tool. "Every app that is built on the platform should help improve the next," Hedin said. One piece of that is Lovable Overflow. It is a store of problems and their fixes, and when a user hits a known issue, a lightweight model pulls the matching fix into the main agent before the user gets stuck.

Stored fixes go stale. A package updates and yesterday's solution becomes today's bug. "Knowledge has a half-life," Hedin said. So the platform tracks how often each saved fix works and prunes the ones that stop helping. Build your own version. Keep a file of the problems you solve and the fixes that work, hand it to your AI when the issue returns, and delete each entry the day it stops working.

The AI tells you when it is stuck

The second idea hands the AI a complaint button. "If you give the agent a tool to tell you when it's feeling frustration, that's another way to do it," Hedin said. When the Lovable agent hits a problem it cannot solve, it files a report to the Lovable team's Slack. A second agent removes duplicates, investigates, and opens a PR (pull request), a proposed code change ready for review. One broken file-copy bug went from agent complaint to merged fix in ten minutes.

The numbers behind these systems are large. Lovable merges around ten fixes a day into production from agent-filed reports. The share of users happy enough to publish their app rose 2%. The stuck rate dropped 5%. "That is on the same order of magnitude […] [of having] a new generation of a foundational model," Hedin said. Give every tool and assistant you use a way to flag what is broken, then fix one thing each week.

What vibe coding at 600 million monthly visits means for you

What does vibe coding at this scale tell you? The barrier to building software is gone. The bottleneck moved from who can build to who knows what to build. Your edge is no longer code, it is choosing the right problem and talking to the people who have it. The platforms winning right now are the ones that get you past the last 10% without a developer. Stop hiding behind the excuse that you can't code and build your first product this week.

Grab my free AI playbook for founders to start building with AI.