At Alma, an AI nutrition coaching app backed by Menlo Ventures, nearly every line of code is now written by AI.
“This is no exaggeration,” said Rami Alhamad, Alma’s co-founder and CEO. “Almost everything we ship today is generated by AI.”
A Business Insider survey of more than 20 startup founders and venture capitalists found that AI is rapidly becoming the leading author of startup code, with Anthropic’s Claude Code being the overwhelming tool of choice. But speed is a double-edged sword, with founders concerned about stagnation and poor quality.
Coding has become one of the most obvious business use cases for generative AI, with VCs pouring billions into AI coding startups like Lovable, Replit, and Cursor. Last week, SpaceX announced it would acquire Cursor for $60 billion. Anthropic has filed paperwork to go public, likely later this year.
According to Dan Lorenc, co-founder and CEO of Chainguard, an open source cybersecurity company, writing AI-generated code is like going from a handsaw to a power tool in woodworking. Powerful but dangerous.
“The AI showed up and handed everyone a circular saw,” he said. “It’s much faster that way, but it’s much easier to lose a finger. Today, everyone is thinking about what guardrails should be put in place to do this safely.”
Lorenc said that 100% of his code is currently written through Claude Code. Last year, he put that number at 60%.
“A year ago, I would have been writing the code myself, but LLM might have saved me some typing time,” he said. “Over the past four to six months, our models, tool calls, and harnesses have gotten much better. We still have to prompt and interact with them, but we can complete tasks in hours to days that previously took weeks to months.”
Volodymyr Giginiak, CTO and co-founder of Wordsmith AI, an AI platform for legal teams, said his company’s code is “almost 100%” generated by AI.
“Very little code is written directly by humans,” Gijniak said. “The distinction is no longer about who writes the code, but how much autonomy the AI has.”
Zyzyniak said fully autonomous tasks currently make up about 10% of jobs, but he expects that proportion to increase rapidly. He predicted that in a year, “80 to 90 percent of tasks will be fully autonomous.”
“Rather than disappearing, engineering is being fundamentally restructured,” he said. “The most leveraged engineers will be those who can design the right environments and contexts for AI to operate in.”
The problem: AI can generate code faster than startups can trust.
In our research, all AI-generated code had clear drawbacks. That’s a lot of slop and bugs.
“The trend I want to warn you about in 2026 is that the ‘vibecoding’ bubble will create a wave of fragile and unmaintainable products made by people who can’t support them after launch,” said futurist and co-founder of SuperTruth and Artists & Robots Jason Alan Snyder.
A December report from Menlo Ventures, an early backer of Anthropic, called this a “cleansing tax.”
“Improvements in the speed of writing code can be offset by time spent on cleanup and quality assurance, an ‘ROI paradox’ that complicates the simple productivity story,” the report states.
At Blueprint, which builds an AI operating system for therapists, nearly all of the company’s code is now written by AI, up from 40% in August, CEO and founder Danny Fried said.
He appreciates that the cost of “trying something” has come down dramatically, but says the value of human employees will only increase.
“Taste and judgment are more important than ever,” he says. “Just because something can be built doesn’t necessarily mean you should build it.”