Token maxing is not AI fluency

Companies rushing to adopt artificial intelligence can fall into an unusual trap: spending money on AI without knowing whether it is actually creating value.

Joey Chan, founder of Cloud Jedi Solutions and a longtime Salesforce community leader in the Philippines, calls one version of this “token maxing.” Some companies encourage employees to use AI as much as possible, even tracking how many tokens they consume, but Chan argues that usage alone says little about whether the technology is helping the business.

“It doesn’t take into account the value that you brought to the business,” Chan said.

For companies moving toward agentic AI it’s definitely worth taking note. AI agents can do more than generate answers. They can work through tasks, use tools, and act on a user’s behalf, making it increasingly important for companies to understand what they are asking AI to accomplish.

AI fluency is more than knowing ChatGPT

Chan said AI adoption should begin with the problem a company is trying to solve rather than simply giving employees access to another AI tool.

“You can’t just use it for non-valuable things,” he said. “Whenever you’re using it, there’s going to be a cost to it. Think about electricity. But in a way, as you use it, it has to really provide value.”

That also changes what it means to be AI fluent. Chan said the skills required depend on a person’s role, but employees need to understand how to give AI enough context to produce useful results and how to judge whether the result is correct.

“Knowing what to ask is actually one of the best skills right now,” he said.

His own experience illustrates the potential. In one recent project, Chan said AI helped his team write scripts for migrating 10 to 15 years of customer data from one CRM platform to Salesforce. A process that could traditionally take around three months was compressed to roughly one to two weeks.

He also used AI to process around 90GB of conference video files, generating session names and details, combining files, and converting them into MP4 format within a day.

Chan said these examples show how AI can allow people to take on work outside their existing expertise. The technology does not eliminate the need for human direction, however. Users still need to understand the goal, provide the right context, and check whether the result is correct.

That becomes even more important as companies move from generative AI toward agents that can perform tasks on their own.

Chan compares an AI agent to an executive assistant. Instead of giving instructions for every individual step, a user can provide a goal and the tools needed to accomplish it. An agent can then determine how to reach that goal, whether that means working through a calendar, email, or other business systems.

The greater capability also means companies need to be more deliberate about how they deploy AI.

For Chan, that starts with training people to use these tools properly and measuring the results in terms of business value rather than usage alone. Companies that buy hundreds of AI subscriptions and encourage employees to maximize their consumption may be able to report high adoption numbers, but that does not necessarily mean they are getting more done.

“The goal is help the business,” Chan said.

That may be a more useful measure of AI fluency than how many tokens an employee can burn through.