As artificial intelligence becomes a standard workplace tool, companies are beginning to monitor how much it actually costs to use. Katherine Bindley reports for the Wall Street Journal that businesses are tracking employee AI consumption through a metric called tokens, the basic unit that measures computing effort behind every AI interaction.
For text-based tasks, roughly 1,000 tokens equal 750 words. Costs rise sharply for code generation, video, audio, or complex multi-step tasks handled by AI agents.
Zapier, an AI automation platform, now runs dashboards tracking token use per employee. Brandon Sammut, Zapier’s chief AI transformation officer, says unusually high token consumption triggers scrutiny. High usage can signal either exceptional productivity or wasteful habits.
At Vercel, a cloud platform startup, one engineer deployed AI agents to build a critical infrastructure service in a single day. The result would have taken a human team weeks. The cost was around 10,000 dollars.
At startup Kumo AI, engineers let AI agents work autonomously, even while the engineers themselves are offline. Co-founder Hema Raghavan notes that high token spending in one area can reduce costs elsewhere, such as by producing more efficient code that lowers cloud bills.
Mark Hull, founder of Exceeds AI, predicts companies will introduce governance rules around token use. He warns that unlimited access creates risk. After granting his entire 15-person team access to an AI coding tool, costs spiked within 48 hours.
