The headline is about AI agents creating code that takes longer to review than to write, calling it an "audit tax."

Short summary

AI code generation is solved, but verification is the new bottleneck: agents write PRs in 90 seconds while humans spend 90 minutes reviewing them. LinearB's 2026 benchmarks show AI PRs take 4.6x longer to review, the real blocker to AI engineering productivity. The fix: three-tier verification (deterministic gates, agent-based review, human approval) plus early evals to measure and reduce the 'audit tax.'

• •AI agents generate code fast but create a verification bottleneck (4.6x longer review time)
• •The 'audit tax' is the time gap between PR generation and merge—the actual measure of AI productivity
• •Solution: deterministic checks + agent-based diff review + human approval + early evals

Generated with AI, which can make mistakes.