Atlassian: Why AI speeds up employees but not organizations

Most companies are failing to realize a return on investment from artificial intelligence because they optimize for individual speed rather than team collaboration, according to Dr. Molly Sands, head of the Teamwork Lab at Atlassian. While 89% of Fortune 1000 executives report that individuals are working faster, only 6% can point to specific examples of clear ROI, as detailed in Atlassian’s annual State of Teams Report.

The State of Teams Report based these figures on a survey of 12,000 global knowledge workers and interviews with approximately 200 Fortune 1000 executives.

Why AI speed fails to translate into ROI

Sands noted that while AI accelerates isolated tasks, it doesn’t automatically improve how work flows across a company. When individuals speed up without a shared direction, they risk crashing into one another, according to Sands.

The report found that roughly 14% of teams have successfully translated AI usage into real value. These high-performing teams differ from the rest of the organization in three specific areas: context, workflows, and culture.

To bridge the gap, Atlassian advocates for a context graph. This system captures goals, decisions, and organizational knowledge in shared digital records using tools like Jira and Confluence, rather than relying on individual memory. This gives AI the organizational context necessary to provide meaningful assistance.

Redesigning workflows and team culture

Winning teams do not simply apply AI to existing tasks. According to Sands, they redesign entire end-to-end processes. This shift prevents the inefficiency that occurs when accelerated individuals operate under different assumptions.

Cultural leadership also plays a role. Sands observed that the fastest-moving teams are led by managers who explicitly encourage experimentation and learning, while accepting that some of these experiments will fail.

Some teams accelerate learning by imposing strict, temporary constraints. Examples cited by Sands include:

• Breaking every task into the smallest practical unit of work, specifically a single story point.
• Committing to write no code by hand for a full week.

Sands clarified that these methods aren’t meant to be sustainable forever but serve as a fast way to learn how to integrate AI into the workflow.

The impact of individual AI ‘hacks’

A significant obstacle to organizational performance is the tendency for employees to figure out AI in isolation. When workers develop their own unique prompts, agents, and assumptions, they create a layer of unspoken knowledge that doesn’t benefit the wider team, Sands argued.

To resolve this, Atlassian experimented with AI working agreements at the start of projects. These agreements require teams to explicitly decide:

• Which tasks AI should be used for.
• Which tasks AI should deliberately avoid.
• Which AI agents will be shared across the team.
• Which common skills are required to keep the team working from the same context.

According to Sands, teams that adopted these formal agreements used AI more frequently, moved faster, made better decisions, and produced higher-quality work.

Sands concluded that AI is not creating new management problems but is instead exposing existing ones. Hidden assumptions and mismatched mental models have always hindered teams; AI simply makes those gaps more consequential.