PwC's chief AI officer isn't impressed by how many agents you have

In the competitive landscape of artificial intelligence, quantifying success can often feel like a numbers game. Recently, Bob Sternfels, CEO of McKinsey & Company, revealed that his firm boasts a workforce of over 60,000, comprising 40,000 human employees and 25,000 AI agents. While this figure might seem impressive, industry experts suggest it doesn't fully capture the effectiveness of a company's AI integration. Dan Priest, the Chief AI Officer at PwC, expressed skepticism about using the sheer number of AI agents as a benchmark for success. In an interview with Business Insider, he remarked, "There was this emerging bragging right around the number of agents I had or I have in production. I think that's probably the wrong measure." Priest advocates for a more nuanced evaluation that prioritizes the quality of AI deployment over the quantity. One of the suggested methods for assessing AI's value is to consider how many agents are recognized as authorities on specific tasks, thereby encouraging human interaction with them. Additionally, evaluating the number of employees utilizing these agents to achieve key business objectives, such as enhancing customer experience in call centers, is crucial. Over the past two years, the focus on AI agents has intensified within corporate discussions about technology adoption. Priest believes that while agents play a vital role in unlocking AI's potential, human oversight remains essential. He pointed out that at PwC, approximately 82% of employees actively engage with AI tools, and the firm diligently monitors agent interactions, accuracy, and the overall impact on operational efficiency. Humans are responsible for reviewing the output from AI agents and providing necessary feedback. "The human is still accountable," Priest clarified. He emphasized that individuals are the ones who achieve certifications and are empowered to make decisions. Initially, PwC and its clients employed a bottom-up strategy for AI adoption, which, according to Priest, yielded disappointing returns. However, transitioning to a top-down approach has proven to be more fruitful, enabling a concentrated focus on fewer agents with deeper expertise in limited tasks. "That agent, I've given them permission to access certain data sets," he explained. "Those permissions are monitored, they expire, they're managed."