Sapna Chadha, vice president for Southeast Asia and South Asia frontier at Google Asia Pacific, described agentic AI as the logical next step in this new technology. “AI agents are where you take intelligent language models and give them access to tools,” she said Tuesday at the Fortune Brainstorm AI Singapore conference. This access allows the language models to stitch together complex and multi-step actions. 

Vivek Luthra, Accenture’s Asia-Pacific data and AI lead, shared one example from Accenture’s own experience: Marketing teams could use an AI agent to manage campaigns, allowing human employees to engage in more value-added functions. (Accenture is a founding partner of Brainstorm AI Singapore)

Chadha predicted that almost a third of all enterprise software will have agentic AI built in by 2028, and could automate almost 15% of day-to-day work and workflows. 

But Luthra suggested that most companies aren’t there yet. Accenture clients fall into three stages of agentic AI adoption. The first is AI assistance, where staff members ask an agentic co-pilot for help in much the same way they might ask a fellow team member a question. The second stage is treating it as an advisor, increasing the overall capability of all human employees and empowering them to make the right decisions. The final stage is giving autonomous agents the authority to handle entire processes on their own.

As of now, Luthra says, most clients are in the first and second stage, with fewer companies prepared to let AI agents truly handle things on their own.

According to Luthra, companies leading the way on AI begin by imagining new ways of structuring workflows, then assess what skills are needed in the workplace to make that happen. Then, they put agentic AI into practice with a cross-platform “workbench” that gives employees opportunities to integrate AI agents into their daily lives.