AI Says “Done.” Who Checks? The $200M Bet on AI Evals

When an AI agent says “done,” who checks? Arena announced a $200 million Series B on October 8, 2026, updated October 9, co-led by Lightspeed Venture Partners and Khosla Ventures, at an announced valuation of $3.1 billion. Rena, Vivian Liu, and Nolan look at AI evals and AI agent evaluation: why a final answer isn't enough, how pairwise rankings differ from checking real workflows, how causal tracing tries to keep comparisons fair, and what the Alignment Index looks for: unauthorized actions, misattributed requests, and false completion reports. Public rankings help you shortlist; your own tasks still decide. Examples and business judgments are our analysis. Chapters 00:00 What comes after “done”? 01:04 Why invest in measurement? 02:09 From preferred answers to completed work 03:14 Comparing real workflows fairly 04:18 Acting and reporting accurately 05:27 Building trust in the platform 06:34 Using the measurement in your business 07:31 What should the financing make visible?