A comprehensive one-year review by McKinsey & Company on the performance of artificial intelligence agents in the workplace has concluded that these digital coworkers require substantial development, are not suitable for every business task, and often fail to meet the expectations of their human counterparts. The findings highlight significant challenges in deploying AI agents effectively at scale.
The consulting firm analyzed over 50 different AI agent implementations it had overseen, revealing that successful integration depends less on the technology itself and more on fundamentally redesigning business workflows. The report suggests that without proper strategy, investment, and oversight, companies risk deploying AI that creates more problems than it solves.
Key Takeaways
- A McKinsey review of 50+ AI agent builds found that they require significant development and management, similar to human employees.
- The report advises against deploying AI agents for their own sake, recommending a focus on reimagining entire workflows to solve specific user problems.
- Low-quality outputs, or "AI slop," can erode user trust and lead to the abandonment of the technology if agents are not properly trained and monitored.
- Effective scaling requires building reusable agent components and implementing robust monitoring systems to catch errors early.
- Human oversight is deemed essential for accuracy, compliance, and handling complex situations, necessitating a collaborative model between people and AI.
A Performance Review for Digital Employees
As businesses increasingly integrate AI agents into their operations, treating them as digital members of the workforce has become common. Following this logic, consultants at McKinsey conducted what they framed as a one-year performance review for these AI systems. The results, detailed in a report by Lareina Yee, Michael Chui, and Roger Roberts, provide a clear-eyed assessment of the current state of agentic AI in business.
The team reviewed a minimum of 50 agentic AI projects they had personally led within the firm and for clients. Their analysis produced six primary lessons for organizations looking to leverage this technology. The overarching conclusion is that deploying AI agents successfully is far more complex than simply plugging in a new tool.
What Are AI Agents?
AI agents are autonomous systems that can perceive their environment, make decisions, and take actions to achieve specific goals. Unlike simple chatbots or automation scripts, they can perform multi-step tasks, reason about problems, and operate with a degree of independence, making them suitable for complex workflows.
Focus on Workflows Not Technology
The first major lesson from the McKinsey report is that companies should stop implementing AI agents simply to have them. The most successful projects were those that used AI to fundamentally rethink and improve an entire workflow, which includes people, processes, and technology.
The authors advise starting with significant pain points for users. For example, organizations with workflows heavy on documentation, such as insurance or legal firms, can benefit greatly from agents that manage tedious, multi-step processes. The report suggests a strategic approach to team building.
"The key question to ask is, 'What is the work to be done and what are the relative talents of each potential team member—or agent—to work together to achieve those goals?'" the McKinsey team advises.
This perspective shifts the focus from a technology-first mindset to a problem-solving one, ensuring that AI is applied where it can deliver the most value.
The Right Tool for the Task
A common pitfall observed was using powerful AI agents for problems that did not require them. If a task is highly standardized, repetitive, and has low variability, simpler solutions are often more effective and efficient. The report notes that for many such cases, companies should stick with established tools.
Simpler Alternatives to Agentic AI
- Rules-Based Automation: Ideal for predictable, repeatable tasks that follow a strict set of instructions.
- Predictive Analytics: Better suited for forecasting trends and outcomes based on historical data.
- Large Language Model (LLM) Prompting: Effective for straightforward content generation or summarization tasks without complex actions.
Choosing an overly complex AI solution can introduce unnecessary costs and complexity. The McKinsey team emphasizes matching the tool to the problem to avoid wasted investment.
Investment in Agent Development is Crucial
One of the most significant issues identified was the prevalence of "AI slop," or low-quality outputs from AI agents. The report highlights systems that look impressive in demonstrations but fail in practical application, frustrating the employees responsible for the actual work.
This poor performance quickly erodes trust, causing users to abandon the technology. To prevent this, McKinsey recommends that companies invest in agent development as they would for human employees. This includes providing agents with clear job descriptions, a structured onboarding process, and a system for continuous feedback to improve their effectiveness over time.
Scaling Demands Monitoring and Reusability
While managing a few AI agents and spotting their errors can be straightforward, the challenge grows exponentially as companies deploy hundreds or thousands of them. According to the report, when a mistake occurs at scale, identifying the root cause becomes incredibly difficult.
The solution is to build monitoring and evaluation directly into the workflow from the beginning. This allows teams to detect errors early, refine the agent's logic, and continuously enhance performance even after deployment. Furthermore, the report warns against creating a unique agent for every single task.
"This can lead to significant redundancy and waste because the same agent can often accomplish different tasks that share many of the same actions—such as ingesting, extracting, searching, and analyzing," the authors noted. Instead, they recommend identifying recurring actions and developing reusable agents and agent components that developers can easily access and deploy across different workflows.
Human Collaboration Remains Essential
Finally, the McKinsey review stresses that AI agents are not meant to replace human workers entirely. The authors emphasize that there will always be a need for people to oversee the process. Human involvement is critical for several key functions:
- Overseeing model accuracy and performance.
- Ensuring compliance with regulations and company policies.
- Applying judgment and context in ambiguous situations.
- Handling edge cases and unexpected scenarios that the AI is not trained for.
The report concludes that for AI agent programs to succeed, work itself must be redesigned to foster effective collaboration between people and AI. Without this focus, even the most technologically advanced systems are at risk of silent failures, compounding errors, and ultimately, rejection by the users they are meant to help.