Does agent quality matter more than workflow redesign?

McKinsey/QuantumBlack, drawing on "more than 50 agentic AI builds we've led at McKinsey, as well as dozens of others in the marketplace," states the lesson plainly: "It's not about the agent; it's about the workflow." The report argues that "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," while organizations that "focus too much on the agent or the agentic tool" end up with "great-looking agents that don't actually end up improving the overall workflow, resulting in underwhelming value."

The mechanism McKinsey describes is process redesign before tool selection: mapping the workflow, identifying user pain points, and judging "how standardized the process should be, how much variance the process needs to handle, and what portions of the work agents are best suited to do." Collaboration between agents and people is then sustained through "learning loops and feedback mechanisms, creating a self-reinforcing system" — illustrated by an unnamed alternative-legal-services provider that logged every user edit in its contract-review document editor, feeding that stream back to "teach the agents, adjust prompt logic, and enrich the knowledge base" so the agents could "codify new expertise" against constantly evolving case law.

This parallels Does easier tool-building actually solve enterprise adoption problems?: both arguments locate the bottleneck away from the artifact itself — Evans at the adoption and problem-definition stage, McKinsey at the workflow-redesign stage — rather than at the sophistication of the tool or agent. It also sits alongside Why do production AI agents stay deliberately simple?, whose finding that real deployments deliberately limit agent scope to preserve reliability is consistent with McKinsey's instruction to fit the agent to the step rather than maximize what the agent itself can do. The companion lesson from the same McKinsey piece, When should you actually deploy an AI agent instead of simpler automation?, operationalizes this workflow-first argument into a concrete decision rule.

The excerpt does not quantify how often workflow-first efforts succeed versus agent-first ones, nor does it give a base rate for the retrenching it mentions in passing ("in some cases, they are even rehiring people where agents have failed"). The evidence is McKinsey's own account of engagements it led or observed, not an independent audit, so the claim should be read as the consultancy's diagnosis of its client base rather than a measured success rate across the market.