Businesses adopt AI agents, but investors still question payback

In short: Big companies are shifting from single AI tools to AI systems that can carry out full work processes, but the cost and risk make the financial payoff hard to judge.

What's going on

Companies are moving away from using one AI model for everything. Instead, they are building systems that send each task to the best tool for the job, like a call center that routes customers to the right specialist. This can make AI cheaper to run and more reliable when used at large scale.

A major change is the rise of “agentic AI”, meaning AI that can take actions, not just write text. These AI agents are starting to run end-to-end workflows in areas like procurement (buying from suppliers), customer support, and software engineering. For example, an agent might check vendor details, collect approvals, and create a report, with limits such as restricted access and “stop rules” that force it to pause if something looks risky.

Software companies are also adding more oversight tools. About half of ERP vendors, meaning makers of large business software used for finance and operations, are launching governance modules. These tools aim to explain why the AI did something and monitor compliance, as governments increase rules and as companies react to high-profile AI mistakes.

What to watch

Investors are still uneasy because it is hard to predict the return on investment, which is a simple way of saying, “Will this pay for itself?” Costs to set up these systems can be high, and companies worry about security, data staying in the right country, and meeting new regulations. With global IT spending expected to pass $6 trillion in 2026, the next test is whether AI agents can save time and reduce errors without creating new risks that are expensive to manage.

Source: Financial Times