According to The Information, pull requests generated by AI agents on GitHub surged from 4 million in September 2025 to over 17 million in March 2026. Cheap generation has rapidly shifted the core bottleneck of software engineering: writing code now takes seconds and costs pennies, while human bandwidth for diff reviews remains entirely static. A Sonar survey confirms the fallout: 61% of developers regularly encounter generated code that looks flawless on the surface but proves to be a ticking time bomb in production.
Venture capital moved quickly to address the bottleneck. On August 12, two automated code verification startups raised nearly $190 million combined: CodeRabbit secured $143 million at a $1.5 billion valuation, while Blacksmith took $45 million at a $550 million valuation. Modern verification infrastructure goes far beyond querying another LLM. CodeRabbit, for instance, spins up isolated virtual machines, runs over 20 static analyzers, and maps live dependency graphs before code reaches human eyes.
For enterprise teams, an unmanaged flood of synthetic code turns into a hidden payroll tax. Without automated screening layers, highly paid senior engineers and tech leads drown in half-baked diffs. The industry is inevitably pivoting to an architecture where automated tooling handles initial filtering, test execution, and architectural audits, leaving human engineers to focus solely on high-level product decisions.