Vibe coding: investors criticize the quality of AI-generated products

What happened?

The term 'vibe coding' has emerged in 2025 to describe the practice of developing software using generative AI assistants, such as GitHub Copilot or Cursor, without a deep understanding of the generated code. While this technique has democratized prototyping, investors and venture capitalists have begun to openly criticize the quality of the resulting products, calling them 'slop'.

According to an article by Sifted, several European investors warn that 'vibe coding' is generating an avalanche of products with serious reliability, security, and maintainability issues. The lack of human review and excessive reliance on AI are creating 'technical debt' that can compromise the long-term viability of startups. The phenomenon is not new: it recalls the 'no-code app' bubble of the late 2010s, when tools like Bubble or Adalo promised to democratize development but ended up generating fragile and hard-to-scale applications. However, the current scale is much larger: it is estimated that more than 40% of new code in early-stage startups is already AI-generated, according to the Stack Overflow 2024 survey. This has led to a situation where investors like Saul Klein (LocalGlobe) have publicly stated that '90% of the products they see in seed rounds have fundamental code quality issues'.

Why is it important?

'Vibe coding' represents a radical shift in how software is created. On one hand, it allows people without technical training to launch products quickly, fostering innovation. On the other hand, the quality of AI-generated code is still inconsistent, and its integration without quality control can lead to catastrophic failures. Investors, who once celebrated speed, now demand technical rigor.

The problem is not just technical, but one of trust. If startup products based on 'vibe coding' fail repeatedly, customers and investors will lose faith in the ecosystem, slowing the adoption of AI in software development. A recent example is the case of FinChat, a personal finance startup that had to shut down in 2024 after its AI-generated code introduced a security vulnerability that exposed data from thousands of users. Moreover, the technical debt generated by 'vibe coding' can be especially dangerous in regulated sectors like healthcare or fintech, where software errors have legal consequences. The European Union is already considering including code audit requirements in the next update of the AI Act, which could directly affect startups that rely on this practice.

Market consequences

• Tightening of investment criteria: VCs are starting to ask about the development process and code quality, not just the idea. Startups that do not demonstrate solid review practices will have more difficulty securing funding. According to Sifted, funds like Seedcamp have already incorporated mandatory 'technical due diligence' into their investment processes, and some have even rejected startups because they could not audit the AI-generated code. This change could reduce the number of funded startups but improve the average quality of products.
• Accelerated technical debt: AI-generated code is often fragile and difficult to modify. Startups that accumulate this debt could face high refactoring costs or even technical collapse. A study by the University of Cambridge (2024) found that the maintenance cost of AI-generated code is 30% higher than human-written code, due to lack of comments and inconsistent patterns. This can be devastating for startups with tight budgets.
• Product failures: Poorly built products can lead to bad user experiences, data leaks, or regulatory non-compliance, damaging companies' reputations. The case of HealthTrack, a telemedicine startup that had to withdraw its app in 2024 due to patient authentication failures, is an example of how 'vibe coding' can have serious consequences. Investors now fear that such failures will become more common and affect market confidence in AI startups.
• Regulation: As AI products become more common, regulators may demand transparency and code audits, affecting startups that rely on 'vibe coding'. The EU AI Act, in its current version, does not explicitly mention the quality of AI-generated code, but legal experts anticipate that upcoming amendments will include 'explainability' and 'auditability' requirements for software. This could increase compliance costs for startups.

What should readers know?

'Vibe coding' is not inherently bad, but it requires responsible use. Developers should treat AI-generated code as a draft that needs review, testing, and documentation. Startups must invest in traditional software engineering practices, such as code reviews, continuous integration, and automated testing. Investors, for their part, should evaluate not only development speed but also product quality and sustainability.

In the words of an investor cited by Sifted:

'Vibe coding can be a powerful tool, but it does not replace human expertise. Startups that treat it as a magic solution are digging their own graves.'

The future of AI-assisted software development will depend on finding the balance between speed and quality. The lesson is clear: AI can accelerate code creation, but human judgment remains irreplaceable to ensure robust and reliable products. As an analogy, today's 'vibe coding' resembles the early days of digital photography, when anyone could take pictures, but quality and composition required skill. Similarly, generative AI is a tool, not a substitute for technical judgment.