When should you actually deploy an AI agent instead of simpler automation?

McKinsey/QuantumBlack argues that "AI agents can do a lot, but they shouldn't necessarily be used for everything," and that leaders should treat agent selection like staffing a team, asking "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 report states that "business problems can often be addressed with simpler automation approaches, such as rules-based automation, predictive analytics, or large language model (LLM) prompting, which can be more reliable than agents out of the box."

The decision rule McKinsey gives turns on two properties of the workflow: variance and standardization. "Low-variance, high-standardization workflows, such as investor onboarding or regulatory disclosures, tend to be tightly governed and follow predictable logic," and here "agents based on nondeterministic LLMs could add more complexity and uncertainty than value." By contrast, "high-variance, low-standardization workflows could benefit significantly from agents" — the report's example is a financial-services company that deployed agents "to extract complex financial information, reducing the amount of human validation required and streamlining workflows," work that demanded "information aggregation, verification checks, and compliance analysis."

This is the operational counterpart to the same piece's broader lesson, Does agent quality matter more than workflow redesign?: where that note states the principle that workflow redesign — not agent sophistication — drives value, this note gives the concrete test for deciding where an agent belongs in that redesigned workflow. It also aligns with Why do production AI agents stay deliberately simple?, whose production survey shows practitioners already constraining agent autonomy to preserve reliability — consistent with McKinsey's advice to reserve agents for workflows that actually need their flexibility. The caution also finds support in Why do AI agents fail at workplace social interaction?, where high-complexity, less-standardized task environments were specifically where agents struggled most, reinforcing McKinsey's warning against defaulting to agents without checking the task's fit.

The excerpt gives no measured figure for how much more reliable simpler automation is "out of the box" compared with agents, and the financial-services example reports only a qualitative reduction ("reducing the amount of human validation required") with no percentage or before/after comparison. This is a heuristic drawn from McKinsey's own consulting engagements rather than an independently tested decision threshold, so it should be read as a prioritization framework for practitioners to apply, not a validated cutoff between when agents help and when they do not.