Asana launches AI Studio, a no-code tool for building AI agents

Asana has launched AI Studio, a no-code tool for building generative AI agents that can automate work planning tasks. 

The company first unveiled its “AI teammate” plans in June, promising to enable customers to build autonomous agents that can perform tasks independently within the work management app.  

On Tuesday, Asana said that the AI agent builder — renamed Asana AI Studio — is now out of beta and available to customers on its Enterprise and Enterprise+ plans in “early access.” There are two options for accessing AI Studio at this stage: a free plan with daily limits on usage, and a paid add-on. (Asana declined to provide specifics on pricing.)

Customers trialing AI Studio during the beta noted several advantages when deploying AI agents, said Paige Costello, head of AI at Asana. “The key benefits we’re seeing are the speed of decision-making and the overall acceleration of work and reduction in administrative and busy work,” she said.

“There is tremendous potential in AI-based agents to expedite workflow,” said Wayne Kurtzman, research vice president covering social, communities and collaboration at IDC. “The ability to deploy agents in the stream of work, where teams work, and without code becomes a powerful proposition.”  

With the launch, Asana also announced additional features for AI Studio. These include a wider variety of potential AI agent actions, more control over smart workflow capabilities such as data access and costs, and an increase in the amount of context the AI agent can reference in its decision-making. 

Users can also view a record of an AI agent’s actions and decisions. “You can actually dig into work that has happened and understand why the custom agent that you’ve built made a specific choice and undo the selection that it’s done,” said Costello. 

Users can choose from four language models to power AI agents: Anthropic’s Claude 3.5 Sonnet and Claude 3 Haiku, and OpenAI’s GPT-4o and GPT-4o mini. 

With AI agents able to complete tasks autonomously, the propensity for language models to “hallucinate” and provide inaccurate outputs could be a concern for businesses.  Costello said there are safeguards in place to help reduce the likelihood of AI agents generating and acting on incorrect information, and argued those designing the AI- workflows are “in the driver’s seat.” 

For example, a user can require an AI agent to seek human approval before carrying out actions deemed higher risk, such as sending an external email to a customer. “People are the decision makers –– they’re the ones ultimately accountable for work,” said Costello.

Adoption of AI agents is at an early stage for most organizations, but it’s accelerating, said Margo Visitacion, vice president and principal analyst at Forrester, covering application development and delivery. Successful deployments will require “experimentation, failing fast, and learning from those experiments,” she said.

“It takes the right level of oversight, focus on the problems you’re solving, and gathering feedback to ensure you’re using the right model that suits your needs,” said Visitacion.