Article illustration 1

Forget theoretical speculations about AI agents – McKinsey just published the industry's first longitudinal performance review of actual digital workers. After implementing and observing over 50 agentic AI systems across real business workflows for 12 months, their team uncovered uncomfortable truths that every tech leader deploying autonomous systems needs to hear.

"Agentic AI efforts that focus on fundamentally reimagining entire workflows – that is, the steps that involve people, processes, and technology – are more likely to deliver a positive outcome," states the report co-authored by Lareina Yee, Michael Chui, and Roger Roberts. This foundational insight sets the stage for six critical lessons learned:

1. Workflow Integration Trumps Standalone Brilliance

Agents deployed as shiny toys inevitably gather dust. True value emerges only when deeply embedded within operational sequences. Document-heavy industries like insurance and legal saw the strongest returns by automating tedious intermediate steps – not just final outputs.

2. Not Every Problem Needs an Agent

McKinsey's blunt assessment: "Approach agents like hiring humans." Before building, ask: "What is the work to be done and what are the relative talents of each potential team member?". For standardized, low-variability tasks? Stick with rules-based automation. Agents introduce unnecessary complexity.

3. "AI Slop" Erodes Trust

Persistent low-quality outputs – dubbed "AI slop" – emerged as a trust killer. Users abandoned agents producing unreliable work. The fix? Treat agents like human hires: "Give clear job descriptions, onboard properly, and provide continual feedback." One-off training doesn't cut it.

4. Observability Becomes Critical at Scale

"When companies roll out hundreds, or even thousands, of agents, the task becomes challenging. When there's a mistake – and there will always be mistakes – it's hard to figure out precisely what went wrong."

Monitoring five agents is manageable; monitoring 500 is a nightmare. McKinsey emphasizes building evaluation directly into workflows with dedicated observability tooling to catch errors before they cascade.

5. Reusability Drives ROI

Organizations bleeding resources build bespoke agents for every micro-task. The breakthrough? Architecting reusable agent components for common actions like data ingestion or analysis. McKinsey notes: "This can lead to significant redundancy and waste" without modular design.

6. Humans Aren't Optional

Autonomy remains a mirage. Human oversight is non-negotiable for model accuracy, compliance, judgment calls, and edge cases. McKinsey warns that without deliberate human-agent collaboration design, systems risk "silent failures, compounding errors, and user rejection."

The verdict? Agentic AI demands more heavy lifting than anticipated – but the potential remains transformative if we abandon magical thinking. As the report concludes, next year's performance reviews depend entirely on whether we heed these hard-earned lessons.

Source: ZDNet - Joe McKendrick