The smartest way to cut AI costs isn’t to use less AI

Two in five AI users have been asked to watch their usage in the past three months. Most kept working exactly as before.

---

You have probably sat through the meeting where someone from finance asks the room to be more careful with AI. The instruction is earnest and the concern is real: enterprise AI bills are climbing, and someone upstairs wants to know what the company is getting for its money. Then the meeting ends and everyone goes back to work, and almost nothing changes.

Atlassian's Teamwork Lab surveyed more than 1,000 US knowledge workers between 25 and 28 August 2026 and found that two in five AI users (40%) say they have been asked in the past three months to watch or justify their usage, or have hit a real limit: a usage cap, lost access or a move to a cheaper tool. Of the workers who felt that squeeze, fewer than a third (31%) actually changed how they work. The rest carried on as before. Another 39% of all AI users say they would resist changing even if explicitly asked to save money.

The scrutiny is real, but it is barely changing behaviour. That gap between instruction and action points to a deeper problem: most people cannot see what their AI costs, and even fewer know how to lower it without simply using AI less.

Visibility is concentrated at the top

Nearly half of AI users (49%) do not know what their company pays for the tools they use, and 46% do not know how to reduce their costs beyond the blunt option of using AI less. The blind spot tracks with seniority. Two-thirds of leaders (67%) can see what AI costs, but that visibility thins fast on the way down: just 37% of managers and 22% of individual contributors can see it. When a mandate for AI efficiency comes down from the top, individual contributors and middle managers are the ones expected to carry it out, yet most cannot see the very spend they would be accountable for managing.

Technical employees fare better: 77% know what their AI use costs, versus 45% of everyone else. One respondent, a manager in government business administration, said: "Most people don't think about it because we never directly feel or see the cost of using AI at work."

The commitment to AI is holding even as cost discipline grows. Eighty-four percent of AI users say their company is holding steady (48%) or going all-in (36%) on AI investment, and just 3% report a pullback. Together those numbers describe cost discipline: companies still betting on AI while they start asking what each dollar returns. The mandate is maturing, from "use more" toward "use it well."

The workers getting the most from AI are the least likely to cut back

Thirty-nine percent of AI users say they would be reluctant to change how they work even if explicitly asked to save money. That reluctance is strongest among the workers who have changed their work the most. Among those for whom AI is "my new normal" or "my team's standard," half (50%) have been told to watch costs, versus under a third (32%) of lighter users. Yet they are the least likely to act on it: among the deep users, 61% say they would be reluctant to change even if it saved the company money.

One vice-president in finance at a tech company put it bluntly: "I don't care how much AI costs, I need it to do my job and the company can pay for it."

The pattern is clearest in marketing, the one function where a majority (66%) report being asked to watch their usage in some way. Dr Molly Sands, head of Atlassian's Teamwork Lab, says the instinct to make your heaviest AI users cut back is backwards. "They're the ones getting the most value from it, which makes them the first people who should get visibility into their spend and the tools to work efficiently, not the first asked to slow down."

What makes AI worth the cost: context, not the model

Asked what makes AI valuable, workers point to the tool they use (42%) and the context they give it (41%) far more than the specific model (13%). A recent Teamwork Lab study found that workers whose AI understood their work were up to six times more likely to call it helpful, and got up to 48% more output they could use without reworking it, compared with workers whose AI had the least context. Better context makes AI more useful and cuts the expensive rework of fixing weak output.

"As models and tools converge, what sets AI apart is how much of your organization it actually understands: your institutional knowledge, your active work, your people, your data," Sands says. "Telling people to 'be careful' with AI mostly bounces off. It's easy to say and almost impossible to act on. When people can see what their AI costs, hone their understanding of where using AI pays off, and have the context to get more from every interaction, the spend takes care of itself."

What this means for L&D and digital transformation leaders

The path to lower AI costs is to make good use easy, not to make access hard. Three moves matter.

First, give people visibility into what their AI use costs. You cannot ask someone to be careful with a number they have never seen. At Atlassian, the company has built every employee a real-time view of their own AI spend, a budget to work within, the freedom to use whatever tool fits the job, and a simple way to request more when the work calls for it. The point is to make cost visible and put people in control of their own usage.

Second, teach people how to get more from every interaction. Beyond sharper prompts, that means giving AI richer organisational context that is connected, relevant and complete, and choosing the right tool for the task instead of defaulting to the most powerful and most expensive one.

Third, put that capability in everyone's hands, not just the budget owners'. The companies that get the most value per dollar will not be the ones using AI least.

"The economics of AI are still taking shape, but the smart move is already clear: help people get more from every dollar rather than asking them to spend less," Sands says. "Let them see what their AI costs, pick the tool that fits, and get the context that makes AI genuinely useful. And put that in everyone's hands, not just the budget owners'. That's what we're doing at Atlassian, and what we'd tell any company to do."

Sources:

• Atlassian, 13 October 2026