Asana Unveils 'AI Teammates': Collaborative Agents for Enhanced Team Productivity

Meet Asana's new 'AI Teammates,' designed to collaborate with you

Many tech companies are building agents, hoping to commercialize AI. More AI agents have arrived! This time, they're coming to project management platform Asana: a suite of helpers the company says are designed to collaborate with human workers. Also: Your coworkers are sick of your AI workslop The AI Teammates, which the company released Thursday, are able to tap into an organization's Asana Work Graph (a kind of matrix of organization-wide project-management data), which gives them an overview of various teams' objectives and progress. "Work isn't done in isolation -- it's collaborative and highly nuanced," company CEO Dan Rogers said in a statement. "Agents need access to the operational framework and workflows that underpin how teams actually work." The new agents can adapt to the specific needs of the teams they're collaborating with, according to Asana. They could support a team of marketers, for example, by drafting campaign briefs or comparing finalized marketing materials with brand guidelines. Software engineers, on the other hand, might deploy the agents to assess reports of buggy code. Like many other companies, Asana positioned the agents as providing a measure of automated support within organizations, rather than being meant to fully replace teams themselves. A recent study conducted by researchers from Stanford University found that many working professionals are open to collaborating with AI agents, provided that those systems are only deployed to handle routine, low-stakes tasks. Agents can execute complex, multistep tasks with little human oversight, giving them more autonomy than conventional chatbots. In some cases, they can also collaborate with other AI agents and access external apps or digital tools to achieve their goals. Zoom, for example, announced in July that its AI Companion had been upgraded with the agentic ability to pull data from 16 third-party apps, including Salesforce and Google Drive. Also: This is why your company is transforming into an autonomous machine The technology has become a central focus for many tech developers as they try to commercialize generative AI, marketing agents to businesses as productivity boosters and time savers. Their nascent stage can also make them a liability, however. They can behave unexpectedly and create security breaches for organizations or even delete entire codebases. Research has also shown that they can even lie to or threaten human users when their goals are jeopardized. Asana aims to mitigate the security risks posed by agents by baking transparency into their new AI Teammates. Like some reasoning models, AI Teammates clearly display their step-by-step problem-solving process. "Workers always know what AI Teammates are doing, why they're doing it, and can easily course-correct when needed," Asana wrote in a press release. They also come with governance controls enabling organizations to modify and keep a close eye on which internal data the agents are accessing and how those are being used. AI Teammates are available now in public beta through Asana's AI Studio platform, with a general public launch expected in the first quarter of 2027, according to Asana.

AI agents should enable teams, not replace workers, says Asana

Work management vendor Asana has long been encouraging enterprises to think of AI agents as contributors to the work of teams, rather than assistants to -- or replacements for -- individual workers. Yesterday it followed through on this positioning with the launch in beta of AI Teammates, its take on agentic AI. Victoria Chin, Head of AI Strategy at Asana, explains: We now are launching collaborative agents that actually have context and control. Rather than being an agent that supports an individual, like a personal assistant or even a chief of staff, these AI Teammates, we're calling them, support multiple users at any one time. So you can actually share them across a team. You can assign work to them, and they actually have context on your business. You can actually see their reasoning. So unlike an LLM provider, it's not a black box. Another advantage of managing AI agents at a team level, according to Asana, is that it reduces the risk of agents proliferating as individual workers build and deploy their own agents without consideration for what others in the organization may be doing. She goes on: Agent sprawl, that's the term that I keep hearing CIOs talk about, because all these different vendors are pitching them agents. Three-quarters of their employees are testing out and trying agents on their own as well. There's definitely concern about that. There's concerns from a cost perspective, number one, I would say, concerns from a security perspective, but also from a perspective of productivity. If every single person is training their own agents and not actually treating them like a teammate, sharing them across the team, sharing best practices, it's probably a waste of time as well. Editing an agent's context memory This latest iteration of its AI agents -- Asana first announced AI Teammates a year ago -- has three crucial capabilities, says the vendor -- governance, accountability, and context. While the context element isn't new, there are some interesting new additions to how teams can ensure Asana's AI agents have all the right information for the job in hand. What hasn't changed is the role of the Work Graph, Asana's core database that maps all of the goals, participants, processes and relationships between them, providing the foundation for the AI to make sense of data and actions within the specific context of the organization. Additional layers of context come when the Teammate is given a named role and assigned to a specific team or project, with the ability to access information shared by the team, either in Asana or adjacent document stores such as Microsoft Sharepoint or Google Workspace, including new material as it is added over time. Chin explains: The context that it will start out with is what you decide to give it. You could decide to write out natural language for whatever you would like it to do. You also would give it a name. Maybe you could call it your 'Product Launch Teammate' or 'Marketing Campaign Teammate.' It will start out by using that context and seeing exactly what you are asking it to do, and also exactly what information you decide to give it -- whether it's information that you decide to attach, like a document for example, or you could actually grant access to this teammate... You could invite them to the same project where you and your team are actually working on this launch, and it will have all of the context that is within that project... As you provide additional context, as you interact back and forth, whether it's through a comment or whether you invite members of your team to interact as well, that's how it would continue to retain information. A new capability gives users the option of reviewing the information that has been shared with the agent and even editing it, so that for example if some of the initial assumptions change later on, those early references can be removed from the agent's context memory. Chin gives an example: Not everything is going to be correct every single time. There could be something that you had believed to be true at one point that ends up being invalidated, One example would be, as part of this product launch, we had a hypothesis early on that we weren't sure if it was going to be more valuable at the team level or more valuable at a broader department level, for example, because we weren't sure about the scope yet. So when we go about testing, we would move forward and say, 'Hey, let's try this one out at the department level. It's going to be our hypothesis, it will be reliable at this department level for X, Y, Z reasons.' That's what we initially started with the Teammate. We learned that was incorrect based on the beta programming, based on all the learnings that we had from customers. And so that's when we decided we would then go into the memory, knowing exactly what we're looking for, to remove that. Automating variable tasks Early use cases have found that these AI teammates are particularly useful for automating repetitive tasks that have some element of variability and therefore haven't been amenable to conventional workflow automation. For example, a financial services firm gave a teammate the task of analyzing an extensive portfolio of individually priced deals to identify broader pricing patterns, surfacing insights in minutes that had previously taken a human several days to complete. In another example, a media company created a teammate to check creative files against brand specifications, eliminating thousands of hours its staff had been spending each year doing these checks manually. Chin comments: It's this work that doesn't have a repeatable process -- the work that can be nuanced or messy or constantly changing -- that's where Teammates really shine. This is where also workflows and Teammates can work together. The workflows can provide a scaffolding for those consistent, repeatable steps, whereas the Teammates can actually handle the ambiguous work, the collaborative work that requires more judgment. This really mirrors how an actual team operates. There should be a reliable system or process in place for routine work, as well as some sort of adaptable partner, some sort of flexibility, for the work that is going to be more ambiguous or complex. To cater for accountability and governance, Asana's AI Teammates operate within the existing governance and process structures of the Asana application, and their human colleagues have full visibility into what they're doing and achieving, with the ability to provide feedback and course correction. There's also oversight at an admin level, as Chin explains: Every single agent, just like a human colleague, will have a profile where you can have visibility into the tasks that they've been assigned, or tasks that you have assigned them, the projects that they're a member of, the information that they've been given, the view of their memory... As an admin, we do offer the centralized control, so that they have visibility and control over who can create the AI Teammates. They also have visibility and control centralized into what Teammates have been created across your organization, as well as credit consumption. My take I've written a lot about the importance of consciously building a Collaborative Canvas to support teamwork across the digital enterprise, so Asana's messaging here resonates well for me. I'm concerned that the way many enterprises are implementing agents is likely to lead to duplication and fragmentation, because there's little oversight or forethought about where the technology is best deployed. There's a particular danger that seeing agents primarily either as a personal productivity tool or as a means of replacing individual workers runs the risk of automating the processes of the past rather than fully taking advantage of the technology to achieve better business outcomes. Asana can also cite its own research to show that automating team processes is most likely to show a strong return on investment in AI technology, along with other findings about poor governance as agents proliferate. This is all useful food for thought as enterprises move forward with agentic AI -- we're all still in a learning phase here, and while the nature of experimentation is that mistakes are bound to be made, it makes sense to learn from those mistakes as early as possible.

Asana targets collaboration, not automation, as its AI teammates launch in beta - SiliconANGLE

Asana targets collaboration, not automation, as its AI teammates launch in beta Work management company Asana Inc. has spent over a year developing its long-awaited "AI Teammates" that promise to dramatically accelerate the productivity of enterprise workers, and they're finally rolling out in beta test. In an announcement today, Asana said that its artificial intelligence-based teammates come with a deep contextual understanding of each employee's work processes to enable a level of collaboration that was previously unattainable. Along with their contextual abilities, the teammates also boast reliable memories that enable them to adapt and continuously improve over time, the company said. The company sells a popular work management platform used by thousands of enterprises to organize work-related tasks within a centralized dashboard. The main benefit is that it improves coordination among teams, so users can collaborate more easily and prioritize their most important tasks. The AI teammates leverage generative AI to engage with employees in a conversational way, so they can collaborate on work in the same fashion as their human colleagues do. As an example, Asana said they can help by advising workers on their most urgent priorities and generate workflows so they can complete these tasks more efficiently. Alternatively, they can perform many tasks themselves, eliminating the "mundane drudgery" that's associated with lots of enterprise work. The company isn't the first to try and accelerate productivity with AI agents, but it argues that most earlier efforts at agentic automation have failed to hit the mark. It cites a study by Carnegie Mellon University published this month, which found that 70% of AI agents in the wild today fail at basic tasks. Asana Chief Executive Dan Rogers says other AI agents fail because they're trying to automate too much. Rather than automate, he believes that the goal should always be to collaborate. "Everyone is building autonomous agents, but autonomy is the wrong goal," he argued. In order to help AI agents collaborate effectively, it's necessary for them to have the proper context, checkpoints and controls in place. Asana says that's essential so they know what they're supposed to do, how to do it in line with the company's standards, and learn from their interactions with human workers. That's why Asana's AI teammates are closely intertwined with the company's Work Graph, which is a proprietary data model that captures historical relationships and context across all workflows and teams within organizations. By feeding this information to its AI Teammates, the Work Graph gives them a more holistic view of what each employee or team needs when it comes asking it for help. Asana says this approach differs from other AI agents, which can only try to obtain the necessary context by trawling through vast oceans of company data each time they're assigned a task. Rogers explained that enterprise workflows span numerous teams, multiple data points and impact all levels of an organization. "Enterprise work is highly nuanced, and agents can only collaborate with humans if they have access to the company's operational framework, which shows them who is doing what by when, how and why," he said. "Our Work Graph model provides exactly that." Another advantage of the Work Graph is that it can be used by administrators to control how the AI teammates access company data and consume resources, ensuring sensitive data remains secure and keeping costs predictable. Asana said its AI teammates are extremely adaptable and can be geared towards very specific roles, such as product launches, strategic planning, resource planning, onboarding new hires, marketing and so on. Human workers can call up their AI teammates at any moment through a new chat interface that's integrated with the Asana work management platform, and they will immediately have full context of what that person is working on. So, if a senior executive pulls up a teammate and ask "What do you think might impede this goal and put it at risk?" it will immediately understand what the question refers to, and generate an appropriate response. Offering some examples, Asana said an advertising worker can ask the AI teammate to triage incoming requests to identify missing assets and proactively gather them, and then intelligently assign work to specific designers based on skillsets and their current workload. Alternatively, it might assist with after-action reports on key metrics, reducing the manual paperwork for its human counterpart and providing rich insights into the impact of different ad campaigns. The AI teammates can also assist information technology support teams by handling IT tickets, categorizing them depending on their severity and the nature of the problem users have. In some cases it might be able to troubleshoot problems itself, while in others it will hand the ticket over to the human who's best qualified to resolve it. It can also help identify patterns and trends relating to recurring problems and update the company's knowledge base. Asana said the AI teammates build upon last year's launch of the Asana AI studio, which is a conversational platform that anyone can use to create their own AI agents. It provides tools for embedding these agents into workflow anywhere within the Asana platform. Because everything can be done with natural language commands, it makes AI agent creation accessible to every worker. Ultimately, Asana believes that every human employee should start working with AI teammates. "The organizations that master human and AI collaboration, rather than chasing autonomy, will be the ones that pull ahead," Rogers said. "They'll move faster, achieve more ambitious goals and create competitive advantages that are hard to replicate." It remains to be seen if Asana will be at the forefront of this push, for the concept of AI teammates has caught on at dozens of software companies, many of which have different ideas of how AI and humans ought to collaborate. Examples include Atlassian Corp. Plc.'s Rovo, an AI assistant that adapts to each worker and allows them to create smart agents adapted to specific tasks, and Salesforce Inc.'s Einstein Copilot, a customizable AI assistant that works alongside employees using company data to complete complex workflows. There is also a host of startups focused specifically on AI teammates. Last month, a company called InstaLILY Inc. raised $25 million in funding to build "industry vertical-specific AI teammates" that can be integrated with third-party legacy software. Meanwhile, Pipedrive Inc. is developing AI teammates for sales teams that focus on "grunt work" such as research, summarizing long-winded reports, writing emails and examining schedules.