What will the future of work look like in organizations where people and AI work together smoothly?
That’s the focus of a new report from Asana’s Work Innovation Lab, “Scaling AI in 2025.”
Asana is a San Francisco-based maker of collaboration software; the company went public in 2020, and created its Work Innovation Lab in 2022.
According to Anna James, Research & Operations Lead at the lab, companies that qualify as “AI scalers” are approaching AI as a new kind of infrastructure that is not optional to build. “They have clear governance in place,” she says. “They have measurement frameworks. They’re talking to their employees and collecting feedback from [them] about what’s working and what isn’t. It’s really taking the more systematic, action-oriented approach, versus just deploying the tools.”
We spoke to James in August 2025, digging into the findings from the company’s “Scaling AI” report, as well as the phenomenon of “digital exhaustion.”
• • •
In the AI scalers report, I found it interesting that a hearty chunk of IT leaders who invested in AI solutions last year (35 percent) actually said they’re pulling back or replacing them with better alternatives. Some even feel like they regret having invested too quickly, while in the past, they regretted moving too slowly. Do you find this to be a valid reaction, and how are the most effective IT leaders resolving this discrepancy?
Overall, what we were seeing at the beginning of this year is a kind of maturation in how leaders think about implementing AI. Last year, it was all about the urgency, the feeling of, “I need to explore the options and demonstrate that I’m being proactive about incorporating this new technology and not getting left behind.” What we’re seeing now is a bit more of that shift from urgency to intentionality. Fast adoption doesn’t always mean that end users are getting the value, and the organizations may not be getting value out of their investments as well.
That must be a difficult balance to strike — thinking methodically, versus making sure you’re actually moving the needle in some way. Is that something that people talk about?
Yeah, I think we looked at it in a couple of different ways in the report as well. On the governance and training sides, you can’t have one without the other. So being able to provide both guardrails, but also the enablement for the end users to actually make use of the new technology, we see much better results when you have both.
When we look at how organizations are measuring the returns on these investments, the majority are strictly looking at efficiency gains or productivity, but we see that there’s a big advantage also of having a sense of whether users are adopting it and what their actual experience with the tools is.
…It’s very hard for IT to be deeply familiar with all of the different applications that they are providing the technology for.
Naturally, most IT leaders are the ones in charge of this AI strategy. but you find that this quarterbacking isn’t enough. What’s it take to break down the siloes and work together, and what comes from it?
Around eight in 10 IT leaders say that they are responsible for driving that AI implementation across the organization. But for a technology that is so widely applicable across different functions, it’s very hard for IT to be deeply familiar with all of the different applications that they are providing the technology for. We’ve been exploring AI councils and cross-functional groups within organizations. The key is having that shared accountability. If IT is the one actually providing and sourcing and sharing these tools, they have to have a partner on the cross-functional side as well to ensure it meets their needs.
One thing that we’ve been talking about a lot is super connectors: employees who more naturally bridge different departments and have that social capital. These individuals are not necessarily executives, but we found that they really play a critical role in driving the adoption across the silos. For executives, to be able to identify those super connectors is going to be highly beneficial.
And not only identify them as they exist now, but maybe higher for them in the future. Now, your report states that AI scalers are 154 percent more likely to follow a centralized deployment model with clear guidelines. In other words, instead of waiting around for good use cases to surface—they identify, codify and scale them on purpose. For organizations that do take this approach, how do they avoid the trap of solving for problems that don’t exist?
I think it really comes back to being able to proactively identify high-impact workflows [and] understand what the most repeatable workflows are. Just in terms of incidents and use, it will increase the likelihood of it being used again and again. It also allows people to get a little bit more comfortable with a specific use case if it is being used very often, and is also more of a rule-based approach, rather than being ad hoc.
Being able to identify across an organization a use case that applies to different functions—project intake, for example, or triage, they’re just universal challenges across different departments—and working really closely with cross-functional partners.
…You really see the best outcomes with the governance and training combined, so organizations need to keep the guardrails in place, as well as increasing employees’ confidence in using the AI solution safely.
It seems like this process requires a real audit of workflows, which feels like both a heavy lift and highly valuable given the long-term benefits of AI. Now, talking about governance, there’s a gap between IT leaders and workers who say their organization has created a centralized AI governance structure. In fact, 32 percent of IT leaders and 15 percent of workers say their organization has created a centralized AI governance structure. Why does this gap exist and how can organizations close it? And how are the most successful organizations treating AI governance in action?
We definitely expect IT leaders to have a higher general baseline awareness compared to individual contributors around the governance structures in place, given their role. But this is just such a significant disconnect that we see across so many different aspects of AI implementation and adoption. Executives are consistently more enthusiastic around the thought of using AI in the workplace. They tend to report higher benefits. They’re using it more often, whereas individual contributors do tend to struggle a little bit more, or at least there’s a bit more of a mix of both enthusiastic and skeptical. They do tend to be less aware of what specific policies are in place, what governance it has been implemented. And that brings me back to the need for both the clear governance enablement, but also the employee training. Again, just one or the other doesn’t really work on its own. You’ll see some benefits, but you really see the best outcomes with the governance and training combined, so organizations need to keep the guardrails in place, as well as increasing employees’ confidence in using the AI solution safely.
I guess training, in and of itself, increases the awareness of this governance too.
It increases awareness. It increases confidence. Last year, we spent quite a bit of time exploring that AI mindset among the employee base, the critical need to understand what that distribution is within the workforce and how to then shift that to be more enthusiastic. That will obviously drive adoption and empower people and inspire people to explore some of those more complex applications as well.
[Digital exhaustion] is something we’ve been looking at for a few years, and we’re definitely not seeing that things are getting any better just yet. I think people hoped AI would be a magic bullet there, but we’re seeing it increase year over year.
I found it interesting that you talked about digital exhaustion. This is something that consumers can relate to, just in all the apps that we have to deal with. Obviously, in the workplace, there’s another layer of tools and applications on top of that. You found that AI delivers the most value when it’s embedded into existing tools. I don’t necessarily think that’s surprising in and of itself, but I think it is interesting that workers at organizations with fully integrated AI solutions are 111 percent more likely to report that AI has reduced their digital exhaustion. So how are the AI scalers operationalizing this?
[Digital exhaustion] is something we’ve been looking at for a few years, and we’re definitely not seeing that things are getting any better just yet. I think people hoped AI would be a magic bullet there, but we’re seeing it increase year over year. We’re seeing other research showing how workers are switching between different apps over 1,200 times a day. For AI scalers, what we see is that integration into existing tools is non-negotiable. AI is being embedded into the platforms that employees are already using, and not just added as a separate technology layer. If you think of workers, everyone’s been given access to ChatGPT or Claude or another tool that is still an additional window that they need to switch between in order to be able to do their work. Being able to prioritize the tools that build AI into the platform for more of that seamless experience, you can really reduce the burden on the individual workers while still seeing the benefits of more productive and efficient outputs. Ideally, this would reduce content switching and lower that digital friction that folks are seeing as well.
I imagine, in addition to the literal time spent switching between windows, there’s the cost of distractions that naturally come from it. Just last week, I copied something that I intended to immediately paste into an email, and as soon as I got to my email, it left my brain. Moving to something I can’t ignore: AI agents. How do you recommend AI non-scalers who are working to turn into scalers approach the AI agent conversation, even if they’re not necessarily at that point yet?
Even among AI scalers, there’s still some work to be done to shift the focus towards AI agent implementation. For non-scalers, the first steps are mainly building literacy. Comprehension was a slightly easier barrier for generative AI—once someone experiences one of the LLMs or the chat-based tools, it feels a little bit closer to what people are used to. For AI agents, only 41 percent of workers can correctly define AI agents at this stage. All of that is just going to put your organization at a disadvantage if there isn’t that clarity.
The other piece that we talk about is starting with global based agents, keeping to those more predictable, low=risk workflows, demonstrating quick wins, immediate value—things like approvals or routing tasks to the right person that just helps organizations build that trust and understanding before moving on to bigger tasks.
One thing that we’re definitely exploring a lot more is the centralized agent library. Not necessarily having a proliferation of unlimited AI agents in the workplace, but being able to use and reuse AI agents more consistently to build that trust across the organization and build a comfort level specifically for those who haven’t used them yet.
The organizations pulling ahead are really investing in enablement and transparency and that cross-functional collaboration, rather than just putting the tools out there…
It seems like organizations have learned their lesson a little bit from jumping into AI without a systemic plan. Maybe when it comes time to implement these agents, it will come with less regrets. With that said, is there anything else you want leaders to take away from these AI scalers insights?
We talked a lot about the organization’s actions and steps that leaders can take. But I think one of the more important things to take away from all the research is understanding that scaling AI is just as much about the people as it is the tools. The organizations pulling ahead are really investing in enablement and transparency and that cross-functional collaboration, rather than just putting the tools out there, making them available, having a policy in place but not necessarily enabling your workforce on it. Taking that systematic and intentional approach will only help the workforce make the most of the new technology.
You must be logged in to post a comment.