AI is writing more code than humans now, and nobody's sure if any of it works.

The Summary

  • Qodo just raised $70M to build verification tools for AI-generated code as production systems fill with machine-written software
  • The bet: code generation is solved, code verification is the actual bottleneck
  • Signal reads as market validation that "ship faster" has hit diminishing returns without "ship correctly"

The Signal

The code generation wars are over, and everyone lost. GitHub Copilot, Cursor, Replit, a dozen startups you forgot existed, they all won the race to write code faster. Developers are shipping 3x more code than three years ago. The problem is that nobody knows if it works until it breaks in production.

Qodo's thesis is clean: as AI agents write more code, the verification layer becomes the constraint. Not "can we write it" but "can we trust it." The $70M round (led by Greenoaks with participation from existing backers) suggests institutional money agrees. This isn't a product announcement dressed as news. This is capital flowing toward the unglamorous work of making sure AI-written functions don't silently corrupt your database at 2am.

The timing matters. We're past the novelty phase of AI coding assistants. Developers who were skeptical two years ago now use them daily. But the trust ceiling is real. You'll let Claude write a utility function. You won't let it refactor your payment processing without serious review. Qodo is building for the gap between those two states: automated verification that scales with automated generation.

The broader pattern: every major AI capability unlock creates a corresponding trust deficit. Language models got better at writing, so now we need better plagiarism detection. Image generation got cheaper, so now we need provenance tracking. Code generation got faster, so now we need automated verification that isn't just "run the tests and pray."

The Implication

If you're building with AI agents that write code, start thinking about verification infrastructure now, not after your first production incident. The companies winning in 2027 won't be the ones with the fastest code generators. They'll be the ones whose generated code ships with machine-verifiable correctness guarantees. Watch for verification tooling to become table stakes in any serious AI development workflow.

Source: TechCrunch AI