AI is changing how you work, but are you learning to use it effectively?

Companies across industries encourage employees to use artificial intelligence (AI) to write, analyse information, automate tasks and improve productivity. But as these tools become part of everyday work, a more important question is whether employees are learning to use AI effectively or simply being encouraged to use it more often.

Recent discussions about workplace AI spending and token consumption have brought this challenge into focus. Amazon has cautioned against “tokenmaxxing” - using AI tokens in large volumes without creating meaningful value. The concern is that higher usage does not automatically translate into better work.

The issue is not limited to Amazon. Several major companies are also reviewing how employees use AI tools and how much they spend on them. Accenture, for instance, reportedly introduced a monthly arrangement under which it centrally funds $29 per employee for Claude usage, with additional spending charged to the relevant project or team budget. The company clarified that this was not a hard limit on employees' access to AI tools.

Microsoft has also asked employees to manage token spending more carefully, cut its internal spending on Anthropic's Claude by more than a third and made OpenAI models the default for internal AI coding. Meta, meanwhile, asked employees to use AI tokens more efficiently after Google restricted its access to Gemini models because of computing capacity constraints.

Uber has taken a more direct approach to controlling costs. After exhausting its annual AI budget within the first four months of 2026, the company introduced a $1,500 monthly spending limit per employee for each AI coding tool, including Cursor and Claude Code. The move also raised a larger question of whether increasing token consumption was translating into more features and measurable productivity gains.

AI training needs to go beyond learning prompts

A July 2026 report by The Conference Board, based on a global survey of nearly 1,300 workers and interviews with 35 enterprise leaders, found that 55% of workers regularly use AI, but only 33% had participated in employer-provided AI training in the previous six months.

The gap is not just about access to tools. Employees need to understand which tasks suit AI, how to frame instructions and why its responses must be checked.

Ankur Dhingra, CEO of ProHance, told Business Today that companies often encourage employees to use AI without enough attention to token efficiency, cost-effectiveness and reducing rework.

“By encouraging consumption, we are asking people to do everything with AI, but not encouraging them to think of token efficiency, cost-effectiveness and reducing rework,” he said.

A study by Nature Human Behaviour offers another perspective. In a randomised trial with 487 management consultants, researchers found employees given ChatGPT access and training performed better on data-science tasks outside their skill sets. However, they found no evidence participants retained the technical knowledge when tested without AI after the experiment. This suggests completing a task with AI does not mean an employee has acquired the underlying skill.

Choosing the right AI model matters

Rashmi Nair, Vice President-HR at OpenText India, told Business Today that using the most advanced AI model for every task can drive up costs without adding much value.

Smaller models can handle tasks like sorting information, extracting data and preparing short summaries. More capable models may be useful for complex analysis and coding. Employees should avoid entering client or personal information into unapproved AI tools and verify important outputs against original documents or sources.

Nair said companies can manage costs by routing simpler requests to lower-cost models and reserving advanced ones for tasks that require them. Better data retrieval systems and fewer duplicate files can reduce unnecessary AI usage.

Companies need to measure outcomes, not usage

Dhingra cautioned against treating token consumption as a measure of employee performance. High usage does not show whether an AI-generated answer is accurate, a task was completed successfully, or time was saved.

Instead, he said companies should assess improvements in work quality, turnaround time, rework and the task complexity employees can handle. He also argued usage benchmarks should differ across departments, since data scientists, product managers and infrastructure teams have different requirements.

A study published in the journal Management Science in February 2026 underlines why workplace AI productivity should be measured through outcomes. Combining three field experiments involving 4,867 software developers, researchers found that AI coding assistants were associated with a 26.08% increase in completed tasks. However, results varied across experiments, with less-experienced developers recording greater productivity gains.

For employees, using AI effectively means knowing when to use it, choosing the right tool and checking its output. For employers, it means providing task-specific training, clear data-use guidelines and sensible cost controls. The real measure of workplace AI adoption should not be how many prompts employees enter or tokens they consume, but whether the technology helps them produce better work.