agentsfactbullishOver the last year, agent evaluations have found that agents are now able to make progress on tasks where success is easily verifiableNarayanan & Kapoor08 Aug 2026https://www.normaltech.ai/p/ai-agents-cant-yet-do-open-ended