You can jumpstart your agentic AI journey

Articulate the vision

Start with a clear vision and strategy for agentic AI integration. The vision should articulate how AI agents will transform the business, desired outcomes, key business goals, and how to start supplying AI agents. Identify areas where agents can address pain points and add the most value, such as routine tasks, complex workflows, or customer interactions.

Understand the spectrum of agent types

Determine the best use cases for specific agent types. The KPMG TACO Framework™—Taskers, Automators, Collaborators, and Orchestrators—helps organizations navigate this evolving landscape. Agent types include:

• Taskers that execute well-defined singular goals with repetitive tasks and minimal complexity.
• Automators that deconstruct more complex goals and manage tasks that span multi-system workflows.
• Collaborators that function as adaptive AI teammates, working interactively with humans to achieve multi-dimensional goals.
• Orchestrators, which are advanced systems designed to coordinate multiple agents, tools, and interdependent workflows at scale.

Launch agentic pilots

Consider where in the organization the technology can provide the most value and how pilot projects should be approached. Target areas with successful generative AI pilots, embed agents extensively within specific functions, or implement them throughout a critical cross-functional value stream.

Prepare to scale agents

Consider whether there are a few AI agent platforms that can be piloted quickly to achieve early wins that can build executive buy-in and momentum. Use tools like the KPMG TACO Framework to identify the types of AI agents that can best address your needs and to determine which departments and processes are most appropriate for agentic AI applications. Eventually, develop a plan to scale to an enterprise-wide strategy, involving multiple agent types working in tandem.

Establish agentic talent performance metrics

Put metrics in place to evaluate the impact of pilot projects, as well as mechanisms for gathering feedback and learnings from stakeholders during the pilot phase. Continuously measure and optimize the performance of AI agents with clear performance metrics to evaluate their effectiveness, including error rates, processing times, and customer satisfaction. Use data and feedback to continuously improve performance, including refining algorithms, updating training data, and implementing new features.

Evolve your Trusted AI governance playbook

Expand your existing AI governance program to account for agentic AI, with mechanisms in place to stay ahead of legal and regulatory considerations. Consider creating a living catalog of all AI agents to track their purposes, dependencies, and performance metrics for effective scaling and governance. Develop robust operational processes to manage and maintain AI agents, including monitoring systems, performance metrics, and continuous improvement methods.

Implement Trusted AI evaluations

Institute a rigorous evaluation and reporting approach for AI agents, as you align your AI governance practices with increasingly stringent regulatory conventions and leading standards. Before launching an agentic AI system into production, conduct thorough testing, which increasingly includes “purple teams” that run rigorous offensive testing and defensive evaluations. Inputs from these processes ultimately inform a trust score, which reflects an AI system’s adherence to the organization’s AI principles.

Institute a robust change management program

Consider the impact of AI on team dynamics, morale, and a culture of collaboration and innovation. Communicate clearly that the integration of AI agents is about enhancing human capabilities, not replacing jobs. Codify the ways you work by dividing human activities into structured and tacit work and systematically capture and codify subject matter experts' tacit knowledge to enable agents' effectiveness. Explore behavioral change approaches that feature role-modeling, peer-to-peer learning, and psychological safety.