In startups, almost every line of code is created with AI – but with the new speed comes an unexpected problem.
At Alma, an AI app for nutrition coaching, almost every line of code is now written by artificial intelligence. “I’m not exaggerating,” says co-founder and CEO Rami Alhamad. “Almost everything we publish is generated by AI.”
The company is not alone with Alma. In a survey of more than two dozen startup founders and venture capital investors, Gründerszene found that AI has quickly become the most important author of startup code. By far the preferred tool: Claude Code from Anthropic.
Billions are flowing into AI programmers
Programming is currently becoming perhaps the most important use case for generative AI. Investors are pouring billions into startups like Lovable, Replit and Cursor.
Last week, SpaceX announced it would take over Cursor for $60 billion. Anthropic has also filed documents for an initial public offering, which is expected later this year.
“AI has given everyone a circular saw”
For Dan Lorenc, co-founder and CEO of cybersecurity company Chainguard, the change is huge.
“AI has given everyone a circular saw,” he says. “You work a lot faster, but it’s also a lot easier to cut off a finger. Right now everyone is trying to figure out what safety precautions are needed to use this technology responsibly.”
Lorenc now has all of his code generated by Claude Code. A year ago the proportion was around 60 percent.
“Back then, you wrote the code yourself, and the language models only did a little typing for you,” he says. “In the last four to six months, the models and development tools have become so good that you can still mainly control them. What used to take weeks or months can now be done in hours or a few days.”
Developers rarely write themselves anymore
The situation is similar with Wordsmith AI, an AI platform for legal teams. “People write very little code themselves anymore,” says CTO and co-founder Volodymyr Giginiak. “The crucial question today is no longer who writes the code, but rather how much autonomy the AI has in doing so.”
According to his estimates, AI currently handles around ten percent of all tasks completely autonomously. In a year it could be 80 to 90 percent.
“Software development is not disappearing – it is being fundamentally reorganized,” he says. “The most valuable developers will be those who create the right context for AI to work effectively.”
The price for the higher speed
But the new speed comes at a price. Many founders report faulty, difficult to maintain or unnecessarily complicated AI code. “The trend I see for 2026: The ‘vibe coding’ bubble will produce a wave of fragile and barely maintainable products developed by people who cannot support them long-term after launch,” says futurologist Jason Alan Snyder.
Investors are already talking about a “cleanup tax”
Alma’s investor Menlo Ventures also warns of the downsides. In a report from last December, the VC firm referred to the phenomenon as the “cleanup tax.”
Some of the productivity gains from writing code could be lost due to the additional effort required for troubleshooting and quality assurance. This leads to a “ROI paradox”: AI makes developers significantly faster, but at the same time creates new work.
Human judgment becomes more important
At Blueprint, a startup for AI software in the healthcare sector, almost all of the code is now written by AI. In August last year this proportion was still 40 percent.
CEO Danny Freed sees this as an advantage: New ideas can now be tested much more cheaply and quickly. Nevertheless, human developers are more valuable than ever. “Taste and judgment are more crucial today than ever before,” says Freed. “Just because something can be built doesn’t mean it should be built.”