KPMG’s latest US AI Quarterly Pulse Survey found that 62% of organizations are using AI agents, with leaders expecting to invest an average of $173 million in AI over the next 12 months. Nearly 6 in 10 organizations report measurable business value from their AI initiatives, reflecting a growing focus on performance monitoring and financial accountability.

“The risk isn’t failing to deploy AI,” says Swami Chandrasekaran, Global Head of AI and Data Labs at KPMG. “It’s deploying AI into yesterday’s operating model and calling that transformation.”An employee may use AI to research or analyze something faster, but if workflows still rely on disconnected systems or long approval chains, any gains are soon negated.

There’s already plenty of evidence that AI improves performance within individual tasks. A study in The Quarterly Journal of Economics of 5,172 customer-support agents found that AI assistance increased issues resolved per hour by 15% on average, with the largest gains among less experienced workers.

Resolving a single issue doesn’t necessarily mean the rest of the process instantly becomes more efficient. In fact, in many cases AI deployment simply means that the workflow remains divided, but the pieces are now automated.

It’s a distinction that should shift the focus of AI deployment from the employee level to the entire workflow. A holistic approach can give leaders a clearer idea of AI’s true potential impact on their business.

KPMG argues that one way to address fragmented workflows is through what it calls the ‘superagent’, a new AI orchestration layer designed to bring together multiple AI agents and systems around a shared outcome. Rather than asking one agent to perform individual parts of a complex job, a superagent can direct specialized agents and bring people in when required. “Superagents matter because enterprise work rarely fits inside one agent,” Chandrasekaran says.

As AI assumes more execution work, human value increasingly shifts toward orchestration, judgment, accountability, and context.

Research shows organizations are realizing value from AI beyond efficiency gains, driving impact across the broader enterprise value chain. The deployment of the superagent is a natural next step in connecting the dots of a truly integrated AI work model.

The conversation is shifting from deploying AI to realizing measurable value from intelligence. But intelligence alone rarely creates value; it must be combined with institutional knowledge, domain expertise, and judgment accumulated within an organization.

To demonstrate and measure the value of AI integration, leaders are now being asked to further their understanding of the whole implementation process.

KPMG recommends focusing on three main buckets. The first is observability, knowing where AI is being used and what for. The second is routing, to match the model or agent to the value and complexity of the work. And the third is attribution, establishing a baseline for success and connecting AI activity to measurable change and business outcomes.

Organizations must also evaluate AI through the lens of resilience, ensuring critical work can continue as models, costs, service level objectives, and technologies evolve.

One way to formalize that discipline, Chandrasekaran argues, is through the creation of an AI Value Office. Reporting into the CEO, KPMG says such an office can help leaders manage AI as a portfolio of investments, rather than a volume of activity, all the while creating accountability for ownership and governance.

“Adoption tells me AI is being used. It doesn’t tell me whether the business is getting better,” Chandrasekaran says. “Leaders increasingly want to know which investments are improving growth, so an AI Value Office can reduce risk and materially change outcomes.”The real challenge is no longer deploying intelligence, but allocating it. Organizations must determine which decisions, workflows, and outcomes warrant expensive intelligence, and where efficient-cost alternatives are sufficient.

The next era of AI maturity won’t be defined by the newest model or by the number of agents an organization deploys. It will be measured by whether businesses can show that high-stakes work is being delivered differently, and whether it was AI that improved outcomes.