AI agents in business: what they can do today and why many projects are stopped

AI agents are the most talked-about technology of the year. Unlike a chatbot, an agent can carry out multi-step tasks, fetch information from different systems and make simpler decisions on its own. But while interest is high, the analyst firm Gartner predicts that a large share of the projects will be canceled. The difference between those who succeed and those who do not is rarely about the technology.

What is an AI agent?

A regular AI assistant answers questions. An AI agent receives a goal and works its way toward it: it reads the material, chooses which tools to use, makes changes in systems and checks the result. One example is an agent that receives an incoming case, looks up the customer in the business system, suggests a reply and creates a task for the right case handler.

What makes agents useful is precisely the connection to your existing systems. Without access to customer records, orders, calendars or documents, the agent becomes an advanced chatbot.

Gartner: over 40 percent of agent projects will be stopped

In June 2025 Gartner published a forecast that more than 40 percent of all agentic AI projects will be canceled before the end of 2027. The reasons are rising costs, unclear business value and inadequate risk controls.^[1] According to Gartner, most projects today are early experiments and proofs of concept driven by hype rather than a clear need.

Gartner also warns about what is called agent washing: vendors rebranding existing chatbots, AI assistants and RPA tools without them having any real agentic capabilities. Of the thousands of vendors marketing agents, Gartner estimates that around 130 offer genuine agentic AI.^[1]

At the same time, usage is growing fast

The forecast does not mean agents are a fad. The same analysis predicts that at least 15 percent of day-to-day work decisions will be made by AI agents by 2028, and that a third of all enterprise software applications will by then include agentic features.^[1] In Sweden, 35 percent of companies with at least ten employees used some form of AI in 2025, an increase of ten percentage points in one year. Among large companies the share was 72 percent and among small ones 31 percent.^[2]

Where agents are most useful today

Agents work best in processes that are recurring, have clear rules and where the information is already digital. Typical examples are:

• Case handling: sorting incoming emails and cases, fetching customer data and suggesting replies.
• Document flows: reading invoices, contracts and forms and entering the information into the right system.
• Compilations: fetching data from several sources and creating reports or decision support.
• Follow-up: making sure things get done and reminding the right person in time.

Five tips for success

1. Start with the process, not the technology. Map how the work is done today before you choose tools. That shows where an agent actually saves time and where a simpler automation is enough.

2. Define a measurable goal. Gartner stresses that the real value comes when AI is used where it delivers measurable improvements in cost, time, quality or scalability.^[1]

3. Solve the integrations early. In MuleSoft's survey of more than 1,050 IT leaders, 95 percent said they struggle to integrate AI into existing processes.^[3] An agent that cannot reach your systems cannot do the job.

4. Keep a human in control. Let the agent suggest and prepare, but let a person approve what matters, such as messages to customers or changes in the finance system.

5. Start small and build on it. Choose a well-defined process, measure the result and scale up only when you see the impact.

Summary

AI agents can remove a lot of manual work, but only when they are built on a well-understood process with clear goals and working integrations. The projects that get stopped are usually the ones that started with the technology. The ones that succeed started with the business.