AI agents at work: real evolution or just hype?
Almost all work management platforms are announcing the same thing this year. They want to stop being a simple tool. They want to become the place where people and AI agents work together, on the same plan and under the same governance.
Asana, for example, is preparing several new applications to automate much of the coordination that eats up our time today: meetings, tickets, client follow-up, product planning.
These are ambitious changes. They are part of a much bigger wave: agentic AI, meaning AI systems that can carry out entire tasks on their own, not just answer questions. This wave promises to move from helping with single tasks to running whole processes without constant human involvement.
We’ve spent more than a decade implementing work management tools at Volcanic. That’s why our question isn’t whether agentic AI will change how we work. That’s already happening. The real question is this: how many of these promises will actually fit a company’s daily operations?
The figure few people say out loud
According to Gartner, companies will cancel more than 40% of agentic AI projects before the end of 2027. The cause won’t be limitations in the AI model. The causes will be different: spiralling costs, unclear business value, or weak risk controls.
The report “The GenAI Divide: State of AI in Business 2025” confirms this trend. It was published by the Project NANDA team at MIT Media Lab. Its main finding is stark: 95% of companies that have tried generative AI see no measurable impact on their results.
This figure isn’t based on guesswork. Between January and June 2025, the research team did thorough work:
- Reviewed more than 300 published AI initiatives.
- Conducted structured interviews with 52 organisations.
- Surveyed 153 senior leaders.
Only 5% of these companies managed to scale their AI projects with real, sustained value over time.
Why most projects fail
These figures don’t mean AI doesn’t work. They say something else, and something more uncomfortable: most organisations aren’t ready to absorb it well.
Projects that fail almost always share the same pattern. And that pattern has nothing to do with which AI model was chosen:
- The team switches on the tool too soon, before properly understanding the workflow it wants to improve.
- Nobody clearly defines which decision the agent should automate, or who’s responsible if something goes wrong.
- The company treats it as an IT project, when in reality it’s an organisational change process that needs support.
We’ve seen this same pattern for years in work management tool rollouts, especially with Asana. Technology is almost never the main obstacle. A weak adoption process is.
What this means for Asana’s new AI features
This pattern gives us useful clues about where to focus first.
Dash is a good example. It’s a layer that adds context: it turns meetings, Slack messages, or emails into structured work. It solves a real, familiar friction point: information getting lost between channels.
The same applies to automations that solve a specific, measurable operational problem, such as Service Management with tickets that have no clear owner. The MIT report itself identifies this type of case as the kind that most easily crosses the “GenAI Divide”: specific problems, with a workflow already defined behind them.
Other features target broader processes:
- Command, built for product and development teams.
- Client Management, built for agencies and teams that deliver work to external clients.
Both have the potential to transform a team’s day-to-day work. That potential is unlocked when a company pairs them with a strong adoption process — the same thing that makes the difference in any AI project.
The AI that actually fits
MIT itself identifies a clear pattern among the 5% of companies that succeed. And it isn’t the pattern you’d expect: they aren’t the companies with the biggest budgets, or the ones that buy the most advanced model on the market.
These companies do three things consistently:
First, they choose a specific problem before a specific tool. They don’t start from “we want AI”. They start from: “this specific task costs us X hours every week, and we know exactly where it gets stuck.”
Second, they build AI into the team’s real workflow. They don’t add it as a separate layer that people have to remember to use.
Third, they support the change with people, not just technology. They define who’s responsible for each decision the agent makes. They train their teams. They leave room to adjust the process while it’s in use, rather than treating it as finished on launch day.
None of these three factors depend on which software provider you choose. At their core, they’re the same principles that separate any digital transformation that works from one that never really gets used. The difference between switching on a tool and transforming how a team works has never been the technology. It’s always been the support around it.
If you’re weighing up how to start using these new AI features in your work management, and you want to try it through Asana, at Volcanic we’ve spent years supporting exactly this process. Let’s talk!