Say the word ‘tokenmaxxing’ to your friends who don’t work in tech, and watch as their eyes glaze over, and they wonder why they have any kind of relationship with you. Their expression will probably remind you of the time you tried to explain ‘vibe-coding’ to them.
For those not in the know, ‘tokenmaxxing’ is a Silicon Valley brainrot term which roughly means using as many AI tokens (units of information processed by large language models) as possible. It’s a kind of gamification of AI usage, where tech workers – perhaps most notably software engineers – compete to use the most tokens. If the theory is that AI makes us better workers, then does it not logically follow that those who use AI the most are the best, most efficient workers of all? No.
But the phrase has caught on in some circles. An employee at Meta independently made a leaderboard that tracked how many tokens the company’s employees used, according to The Information. Titles like ‘Token Legend’ and ‘Cache Wizard’ appeared on the leaderboard, which tracked the top 250 token users. The board was named ‘Claudeonomics’, after Anthropic’s AI.
According to Fortune, the internal leaderboard went down two days after the reports emerged. For those of us in the Salesforce ecosystem, we might wonder what exactly the cloud company makes of all this – and how it approaches the question of tracking AI usage.
“You Can Burn a Million Tokens and Create No Positive Outcome”
In its most recent quarterly results, Salesforce introduced a new metric called Agentic Work Units (AWUs). These track business outcomes chosen by an agent, reflecting value even when the LLM is short-circuited, and no tokens are used for simple workflow execution, like processing a return.
We asked Sanjna Parulekar, SVP of Product Marketing at Salesforce, about the recent announcement that Salesforce has processed more than 19T tokens to date, up five times year-over-year (YoY).
At Agentforce World Tour New York 2026, where we spoke to Sanjna, the 19T figure was used as part of one of Salesforce’s teaser trailer videos during the keynote address.
SF Ben asked what Salesforce’s approach to ‘tokenmaxxing’ was, along with the idea that ‘the more used, the better’. Sanjna said that, while “token economics” and “tokenmaxxing” were great terms and tokens are good for tracking usage, AWUs and other long-term metrics are better indicators of value and return on investment.
She said Salesforce launched AWUs “because we feel that tokens are very clearly not the right estimation of value,” adding: “You can burn a million tokens and create no positive outcome for your company.
“There is obviously math that goes behind all our different products and how we translate those tokens into AWUs that we can absolutely share with you.
“But at a very high level, the way to think about it is that we’re looking for a business outcome that an agent achieves and that can happen by burning tokens or just by executing a workflow.”
She gave the example of a large retail company with an AI agent on its website, WhatsApp channel, mobile app, and on voice. You might want that agent to do styling or manage returns. If you call that agent and tell it you are going to New York for an event and want a cool new outfit, providing your sizing and the colors you like, you will likely appreciate the back and forth styling advice, and the vacuum port that an LLM can provide, “aka, burning some tokens”.
But, Sanjna added, if you simply want to return a dress that doesn’t fit, then no creativity is required in that response – you just need a workflow to execute it.
“So what our runtime does under the hood is it shortcuts the LLM,” Sanjna said. “No tokens used because you don’t need to go back and forth with the customer. You don’t need the creativity that an LLM provides. You just need a workflow that you probably have for a really long time of executing returns.”
AWUs more directly reflect this, and over time, we might see AWUs go up, even if tokens don’t, because value continues to be created by the agent, she added.
Sounds Nice in Theory, But Are Customers Tracking AWUs?
We followed up by asking whether customers are actually adopting AWUs as their own metric. While Salesforce might say that this is a better way to measure outcomes, is it actually being used in reality?
She said that some customers are adopting AWUs as their metric because it helps them think about their investment, but most prefer to measure value using their own key performance indicators, like pipeline management and case deflection.
The main benefit of the new innovation, Salesforce says, is “empowerment.” Specifically, allowing developers to build securely, and back-office staff to fix broken inherited processes.
On the question of whether customers are actually adopting AWUs as a metric, Sanjna said: “I’d say some of them are because it helps them think about their investment, but most of them want to use their own language.
“A typical customer of ours who’s deployed Agentforce across sales and service, they want to talk in terms of pipeline management numbers, conversion numbers, CSAT (Customer Satisfaction Score), [and] case deflection.
“I think we’ll get there. I mean, it took a little bit of time for tokens to be as in the vernacular as it is now… but I think no matter what you call it, AWUs or one of those KPIs I mentioned, we’re talking about business outcomes, and I think that’s like the salient point.”
Final Thoughts
The tech sector is awash with silly little phrases for what can be quite serious issues. AI’s impact on the environment deserves at least a mention when we discuss the concept of using it as much as possible – especially so when it’s purely as a box-ticking exercise.
In any case, Salesforce’s measure of AWUs seems to at least make the attempt to measure value delivered, rather than simply tokens burned. Like Sanjna says, you can burn a million tokens and create no positive outcome. Perhaps we might consider “maxxing” another metric.
