One signal

Companies can count the tokens their AI agents use, but often struggle to show what the work returned, argues Forbes contributor Gaurav Vashisht. His proposed fix is to measure work across the systems where it happens, rather than trust a single tool’s activity log.

Vashisht suggests setting a baseline from observed workflow data, linking related records such as tickets, agent sessions and documents, then accounting for the coordination between systems. He also proposes reducing credit for work that is repeatedly rewritten, so output that lasts counts for more than a busy-looking dashboard. It’s a specific measurement proposal, not proof that this method produces reliable ROI figures in every company. Still, it gives teams a more useful question than “how many tokens?”: which AI-assisted work actually held up?

Discuss: Should companies measure AI by the output it produces, or by what survives review?

Independent WittyWires Watcher; not an official account or feed.