AI is making employees faster, but why isn't it making companies much more productive?

Artificial intelligence is helping employees write faster, analyse information, generate code and complete routine tasks in less time. Yet for many companies, that improvement has not translated into a similar jump in overall business performance.

That is emerging as one of the biggest questions around enterprise AI. Companies have spent heavily on tools and are encouraging employees to use them, but measuring what those gains mean for revenue, efficiency, quality or customer outcomes remains difficult.

Research from Atlassian's Teamwork Lab found that 89% of executives surveyed said AI had increased the speed of work, but only 6% said they were confident they could point to clear examples of organisation-wide AI return on investment. The research surveyed 12,035 knowledge workers and 173 Fortune 1000 executives in January and February 2026.


For Avani Prabhakar, Atlassian's Chief People and AI Enablement Officer, the disconnect comes from treating AI as a tool for individual productivity rather than as a change to the way teams operate.

“Everyone is building their own agents for their workflows,” Prabhakar said. While that can make individual employees more productive, she argued that the gains do not automatically carry over to the wider organisation.

When everyone moves faster, coordination becomes harder

Consider a software team in which developers use AI to generate more code, product managers use it to prepare documents and analysts use it to produce reports. Each person may finish their own work faster. But the work still has to move through reviews, approvals, meetings, testing and decision-making.

If those parts of the organisation do not speed up at the same time, the additional output can create another bottleneck. Atlassian describes this as an “AI fragmentation tax”. Its 2026 research estimates that the cost of fragmented AI-driven work could amount to $161 billion a year for Fortune 500 companies. The company says 87% of knowledge workers surveyed reported that they lacked the time or capacity to coordinate as more people moved into execution mode.

Prabhakar said this is why the difference between individual productivity and organisational transformation matters.“Individual productivity gains from AI adoption are running at around 33%, while transformational gains are stuck at 3-4%,” she said.

The figures are based on Atlassian's internal observations, rather than a universal measure of AI productivity across companies. But the underlying problem is broader: making one task faster is easier than redesigning an entire workflow.

The ROI question is changing

Early corporate AI programmes often focused on adoption. How many employees were using the technology? How frequently were they using it? Which tools were most popular? Those numbers are relatively easy to measure.

The harder questions come later. Did the customer receive a faster response? Did a product reach the market sooner? Did the quality of decisions improve? Did the company reduce costs without creating additional risks?

“Companies need to move beyond measuring usage and speed. It isn't a typical tech transformation where you automate a process, and then you are done,” Prabhakar said. “The bottlenecks come down to how teams are structured and how people work together.”

More AI does not automatically mean more output

There is another complication. Employees do not necessarily use AI in the same way. An engineer may use an AI assistant to write and test code. A salesperson could use one to prepare customer information. An HR professional might use AI to summarise documents. Applying one productivity metric to all three roles can produce misleading results.

“The real differentiator is your team's context,” Prabhakar said. She described an internal test in which the same prompt was run first against a blank model and then against Atlassian's internal knowledge graph. According to Prabhakar, the version using company context produced a better answer faster while reducing token costs by 48%.

The broader point is that access to a powerful model is only one part of the equation. The quality of the company's data, knowledge systems and workflows can influence what employees actually get from AI. Atlassian's research similarly found that only 22% of surveyed knowledge workers fully trusted AI's accuracy, while 69% said their data and knowledge foundations were not optimised for AI.

AI is changing the definition of productivity

The next phase of enterprise AI may therefore involve fewer questions about how many employees are using AI and more about what happens to work after adoption. A company may discover that an employee can produce twice as much content with AI. That does not necessarily mean the company needs twice as much content.

The value comes when the additional capacity is used to solve a customer problem, make a better decision, launch a product sooner or eliminate unnecessary work.

Why AI may increase the need for collaboration

The debate around AI and jobs has often focused on whether machines will replace individual workers. The productivity question points to a different possibility. If AI takes over more repetitive work, employees could spend more time on coordination, problem-solving, customer interaction and decisions that require judgement.

That could make collaboration more important, rather than less. Atlassian's research found that only 24% of leaders currently focus their AI efforts on improving teamwork, even though the company says around 80% of work happens at the team level.

For companies, that means the next AI investment may not necessarily be another model or another chatbot. It could involve redesigning workflows, improving internal knowledge systems, establishing rules for AI use and deciding where humans remain responsible for final decisions.