Move over, tokenmaxxing.
Focusing on AI usage rather than what it delivers risks incentivizing the wrong efforts, said Laura Gonzalez, head of people at the AI video company Synthesia.
The better goal, she said, is impact. Her view aligns with a sentiment from a growing number of leaders that heavy AI use isn’t necessarily a win for business.
Measuring software developers chiefly on token usage would be like evaluating a salesperson based on how many calls they made rather than how many deals they closed, she said.
“It’s not the metric for success,” Gonzalez said of token use. Instead, she said, companies should focus on the business impact AI delivers.
Rather than looking at how many tokens someone burns, a common metric for measuring AI usage, better questions include whether the tech is helping teams ship faster, prepare better for customer calls, or shorten the time it takes new hires to start making an impact, Gonzalez said.
Synthesia, based in London, develops software that lets businesses create AI-generated videos for training, marketing, and other purposes, so its workers use AI broadly across all roles, she said.
While the company doesn’t reward workers for going wild with token use, it still wants to see them use the tech.
“We expect absolutely everyone to leverage and use AI when necessary internally,” Gonzalez said.
A shifting tide
Not focusing on tokens marks a departure from the messaging at some companies, where leaders treated burn rates as an indicator of AI adoption. In some cases, leaderboards highlight employees generating the most AI activity.
Now, surging AI costs for some heavy users are prompting some bosses to rethink what they tell workers.
At Coinbase, AI usage spiked in February after Claude rolled out a new coding model. Now, the crypto exchange uses price caps from $500 to $5,000 a week, depending on a worker’s level and job, Business Insider previously reported.
Marty Kausas, CEO of the customer-operations platform Pylon, said Wednesday that his company’s annual Anthropic bill is set to jump from $400,000 to $1.4 million as it moves to an enterprise pricing tier. Kausas, who said he inadvertently spent $4,000 on Claude Code in three days, predicted that the era of tokenmaxxing is “coming to an end.”
Gonzalez said cost concerns don’t drive the company’s approach to deemphasizing tokens. In January, Synthesia announced it had raised a $200 million series E round, valuing the firm at $4 billion.
Not encouraging going big on token use also doesn’t mean forgoing innovation, she said.
“You can still have a culture of experimentation and providing people with the tools to go run with things without rewarding token maximization,” Gonzalez said.
When tokens matter
There are still times, like during a hackathon, when the company has used a leaderboard to show token use. Yet those are the exceptions, she said.
The corporate push-pull over token use comes as many employers seek ways to incentivize AI use among their workers to boost productivity. In some cases, managers are using dashboards to track how much staff in roles like software development are using AI. In some cases, employers are considering AI use in how they evaluate workers.
Token allocations can still matter, particularly for new hires who might arrive with different expectations about access to AI tools.
Prospective employees might still want to understand which AI resources they’ll have access to, said Jeff Hyman, a recruiter with about 30 years of experience.
He compared asking about token allocations to a question workers routinely asked in an earlier era: “Does the job include a laptop?”
Do you have a story to share about your token use? Contact this reporter at [email protected].
