By now most people have heard of AI agents — software that can act autonomously to undertake a series of tasks, but Asana has decided to take a different approach when it comes to AI. The company on Wednesday introduced a beta of what it’s calling ‘AI teammates,’ in a bid to help move work inside an organization.
Paige Costello, head of AI at Asana, said the company chose the name deliberately to create a mental shift in terms of how people think of interacting with AI at work. “We believe that the future of work is humans not just working with humans, but humans also working with AI,” Costello told TechCrunch.
“And we believe in that world that it’s going to be just as important to understand what you asked the AI to do, what it did and how much it cost to make that happen.”
Costello said it’s about creating transparency and structure around the AI so that businesses can specify and create customized assistants to execute core parts of workflows.
It sounds good, but what does this look like in practice? According to Costello, the previous generation of workflow tools were rigidly defined, and what separates today’s announcement (and generative AI in general) is that it provides a more flexible way to move work inside a company.
So as work comes in, the AI could evaluate the current state and determine if it’s ready to move to the next step, or if it needs to return the work to a human to add more information before it can continue. For example, if a help ticket comes in with missing or inadequate description, the AI teammate could send it back to the person who submitted the ticket, asking for what it needs. That could involve using generative AI to help the human employee write the ticket before sending it on to the AI teammate, which can then direct the ticket to the right person for resolution.
Asana clearly has a treasure trove of data related to how work moves inside a company to train its models, thanks to its work graph, which models how work is connected between individuals and departments. But while this all sounds good, we know that AI agents can still hallucinate, and they don’t always understand the nature of an activity.
“The work graph enables us to tell AI not just how work happens, but how work happens in this specific instance. So when we embed AI teammates into a particular workflow, they’re given a specific job to do. When they have that specific job, and they know what information to read, they’re much more likely to do the right things,” Costello said.
But Costello acknowledged that they are encouraging customers to keep humans involved, because they recognize that the AI isn’t always going to get it right. “I would say that a core principle we have about AI at Asana is ‘human in the loop.’ We believe that, ultimately, humans are responsible for decisions and they’re accountable for results,” she said.
That means humans need to be able to supervise and inspect the AI to make sure it’s making sound recommendations in line with the company’s values and way of working.
To solve for this, Asana has been looking for workflows where they’ve been able to achieve a high degree of accuracy. “We have found that we’re able to embed AI teammates to remove a lot of administrative work and tracking work within these systems very quickly, with high degrees of success. The other thing we’re finding is that we’re effectively able to use dynamic variables to retrieve information about the work and about the systems in the context of work that is also very good,” she said.
All that said, this is still in beta, and it will likely involve some growing pains, especially when companies try to move beyond experimentation and implement it at scale. But if data is the key to building more accurate models, an organization with insight into how companies work, like Asana does, could have a better chance than most of successfully helping move work through a series of steps in a more intelligent way.