AI's productivity promise has a math problem

Workers using generative AI report a 33% productivity boost, but only 3% of organizations say they're seeing real business transformation. Someone's math isn't adding up.

Atlassian's AI evangelist Sven Peters has a theory about why.

"We're investing a lot in AI — we're doing a lot, but we're stopping at individual productivity,” said Peters, who spoke to Nextgov/FCW ahead of his March 2 talk at Talent Arena, an international conference dedicated to digital talent held in Barcelona. “We're not taking the next step." 

The numbers come from Atlassian's 2024 “AI Collaboration Report,” but Peters, who works with organizations across all industries, says the pattern holds everywhere he looks. Most organizations treat AI adoption as finished once employees are using it at their desks. Emails get summarized. Documents get drafted. Code gets written. Then everyone wonders why the business isn't transforming.

"You can't just screw AI on everything — it only makes you faster," he said. "It means you need to think about, ‘how are our teams collaborating? How are people collaborating?’ You probably need to change the way you work."

The harder question — where does work actually stall? — is the one most organizations aren't asking. Peters' answer: Look for the handoffs, the moments between teams where projects slow down, stop and restart without anyone owning the gap. Applying AI to the wrong phase produces nothing. 

For example, if code review is the chokepoint in a software pipeline, writing more code faster just deepens the backlog on the other side. The fix, Peters says, starts with finding where work actually breaks down — and applying AI there. That's the hard work most organizations skip.

Silos make it harder. Departments operate independently, handoffs break down, work stops without explanation. It's dysfunction that predates AI by decades."

One department doesn't know what the other one does," he said. "That's where it cracks."

The answer is mapping where value actually flows and pointing AI at the friction, not layering it on top of workflows that were already broken.

The near-term upside is in agentic workflows, where AI handles multiple steps without waiting on a human at each handoff. At Atlassian, that's already showing up in code reviews, customer feedback loops and onboarding. Legal and HR are among the heaviest users.

"I want to see AI applied to real companies' workflows and helping real organizations to thrive," he said. "Not the demo that looked great on stage."

Getting there fast and getting there responsibly aren't separate conversations. Large language model developers, platform vendors and end users all share it, Peters said.

"We have a responsibility as LLM vendors to build ethical AI, but then the vendors like Atlassian that use those LLMs have the responsibility to make AI correct," he said. "But also the users of AI technology — you can misuse it. I think we share all this responsibility when it comes down to using AI in an ethical way."

The next two to three years, he argues, will be the period when organizations stop declaring victory after handing employees a new tool and start doing the harder work. That includes opening knowledge bases across departments so AI can draw connections across silos, not just within them and rebuilding workflows around human-AI collaboration.

But for all the talk about speed and transformation, Peters keeps coming back to something simpler."

We still are humans," he said. "We still need to collaborate. We still need to work together. That's not going away with AI."