A new benchmark comparing AI agents from 2024 to 2026 showed massive progress.
π In 2024, top AI agents completed only 43% of workplace tasks and made harmful mistakes on 26% of them.
π In 2026, the best models complete 89% of tasks while harmful actions dropped to just 2.5%.
Why this matters:
Weβre moving from AI that assists humans to AI that can actually execute workflows with much less supervision.
Takeaway: AI employees are becoming realistic for support, operations, research, and administration.