Autonomous AI agents fail at 70% of basic workplace tasks, and 84% of enterprises cite governance risk as the barrier. Asana answer: agents that live inside work management infrastructure, not standalone chat. The catch: the beta numbers come from the already-converted.
image from grok
When Asana's chief product officer Arnab Bose says autonomy is the wrong goal for enterprise AI agents, he has data that suggests the contrarian framing isn't just positioning. Research from Carnegie Mellon University shows autonomous agents fail at roughly 70 percent of basic workplace tasks. Asana's answer is a new class of agents that live inside work management infrastructure rather than standalone chat interfaces, now in beta with a Q1 FY27 general availability target, available to select enterprise customers as a paid add-on to the existing platform.
The product, called AI Teammates, puts agents inside Asana's Work Graph, the platform's data model that maps every task, project, team member, and dependency across an organization. Instead of an agent operating in isolation with no visibility into which approvals are pending or who owns what, AI Teammates read the same structural context that human teammates already use. The feature includes 21 off-the-shelf agents for marketing, IT, and operations roles, with a no-code builder for custom deployments. The design principle Asana calls "multiplayer by design" also means agents can connect to third-party applications via bi-directional API sync with Google Drive and Microsoft 365, with HubSpot and Salesforce connectors in development. MCP connectors for more non-deterministic agent-to-agent interactions are planned for next quarter, starting with Slack.
"We believe in AI being 'multiplayer' by design," Bose said in an interview with Computerworld. "The future of the agentic enterprise will only be realized if agents can work independently and with multiple people, versus just a copilot."
The production gap Asana is targeting is well-documented. According to Camunda's 2026 State of Agentic Orchestration report, 71 percent of organizations say they use AI agents, but only 11 percent of agentic AI use cases reached production in the last year. The same survey found that 80 percent of organizations report most of their AI agents are currently limited to chatbots handling summarization or Q&A rather than mission-critical workflows. Eighty-four percent cite business risk when IT lacks appropriate controls, 80 percent point to insufficient transparency around how AI operates in business processes, and 66 percent flag compliance concerns.
Those are significant numbers, but Camunda is a workflow orchestration vendor with a commercial interest in the "agents need better orchestration" conclusion. Treat the survey as industry self-reporting, not independent market research.
The beta data Asana published from its own program is more attributable. Across more than 200 participating organizations, 93 percent of AI Teammates were granted full edit access rather than view-or-comment-only permissions, which the company reads as evidence of trust. Teams in the beta finished work twice as fast, and tasks managed by AI Teammates were 3.2 times more likely to have a clear owner and 2.6 times more likely to have a defined deadline. The limitation is the obvious one: beta participants were self-selected Asana customers with high baseline workflow maturity, so the numbers reflect enthusiasm from the already-converted.
What makes the competitive picture more complex is the so-called SaaSpocalypse scenario, the theoretical case that general-purpose AI agents become capable enough to operate SaaS applications directly, making the work management layer underneath redundant. Bose's counterargument is that Asana still benefits even in that world: if a user's agent pulls data from Asana to get organizational context, it makes the Work Graph stickier, regardless of which agent is doing the pulling.
Craig Le Clair, VP and principal analyst at Forrester, gives Asana credit for a genuine data advantage but sees the real competitive threat coming from horizontal platforms rather than pure-play work management vendors. "Asana is well-positioned against other pure-play collaborative work management vendors," Le Clair told Computerworld, "but the real threat are these more general alternatives," specifically Microsoft 365 Copilot and apps like Teams, and Salesforce with Agentforce and Slack.
That's the architecture question the beta data can't answer. If the route to reliable agents runs through existing workflow infrastructure rather than more capable models, it changes what enterprise buyers should be evaluating. But if Microsoft's agent layer becomes good enough, the Work Graph advantage shrinks considerably. Asana is betting that work management data is durable infrastructure regardless of which agent sits on top of it, a bet that depends heavily on how fast Microsoft and Salesforce close the gap.
Pricing for AI Teammates is $15 per user per month for 100 agent requests, with overage at the same rate per additional 100 requests, according to Fast Company.
The structural problem is real and documented: 11 percent production adoption despite 71 percent usage. Whether Asana's specific answer, a Work Graph as agent substrate, is the one that survives contact with Microsoft and Salesforce's agent strategies is the open question worth tracking.
Asana's announcement is at investors.asana.com. Camunda's 2026 State of Agentic Orchestration report is at camunda.com.
Story entered the newsroom
Research completed — 0 sources registered. Asana CPO Arnab Bose argues enterprise AI agents should be multiplayer by design — embedded in shared workflows, not private chat. The genuine find: 9
Draft (915 words)
Reporter revised draft (775 words)
Reporter revised draft based on fact-check feedback
Reporter revised draft based on fact-check feedback
Reporter revised draft based on editorial feedback
Approved for publication
Headline selected: Your AI Teammate Shouldn't Act Alone
Published (787 words)
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Agentics · 3h 46m ago · 4 min read
Agentics · 4h 18m ago · 4 min read
