QuantumBlack publishes enterprise agents playbook
- McKinsey’s QuantumBlack published a June 2025 playbook arguing companies should treat AI agents as workflow operators, not chatbot add-ons, to get results. - The report says agents need modular, observable infrastructure and human-centered operating models, while newer McKinsey research says fewer than 10% scale. - McKinsey’s 2026 follow-up ties agent rollouts to data quality, governance, and auditability rather than more pilots. (mckinsey.com)
AI agents are software systems that can plan, call tools, and take actions, not just answer prompts. McKinsey’s QuantumBlack says that difference is what makes them useful inside real workflows. (mckinsey.com) In its June 2025 report, “Seizing the agentic AI advantage,” QuantumBlack called agents a way to move generative artificial intelligence from a sidecar assistant to a goal-driven collaborator embedded in core processes. The report was written by Alexander Sukharevsky, Dave Kerr, Klemens Hjartar, Lari Hämäläinen, Stéphane Bout, Vito Di Leo, and Guillaume Dagorret. (mckinsey.com) The playbook’s core prescription is architectural: use open, extensible, observable infrastructure so multiple agents can work across systems and still remain under human control. It says companies need modular and resilient design before they can trust agents with production work. (mckinsey.com) McKinsey sharpened that message on April 2, 2026, with a separate article saying nearly two-thirds of enterprises have experimented with agents, but fewer than 10 percent have scaled them to deliver tangible value. Eight in ten companies cited data limitations as a roadblock. (mckinsey.com) That framing helps explain why QuantumBlack keeps emphasizing plumbing over demos. If agents are going to make decisions and execute tasks at machine speed, the company argues, they need clean data, access controls, lineage, and traceability. (mckinsey.com 1) (mckinsey.com 2) McKinsey’s banking examples show the target use case: automate whole chains of work rather than isolated tasks. In frontline sales, the firm says banks that rewire one domain end to end have seen 3 to 15 percent higher revenue per relationship manager and 20 to 40 percent lower cost to serve. (mckinsey.com) The same logic appears in McKinsey’s banking operations research from February 27, 2026. It says banks that move beyond pilots can use agentic systems for less structured, personalized work that older automation tools struggled to handle. (mckinsey.com) Governance is the other half of the playbook. In a March 5, 2026 McKinsey podcast, partner Rich Isenberg said “agency” is a transfer of decision rights, which shifts the question from model accuracy to accountability when software acts. (mckinsey.com) That is why observability keeps recurring in McKinsey’s agent guidance. The firm says the hardest failures are the ones companies cannot reconstruct afterward because they did not log the workflow, ownership, and scope of what agents were allowed to do. (mckinsey.com) McKinsey is also pushing agents into software modernization. Its QuantumBlack-backed LegacyX offering uses specialized agents to handle end-to-end legacy information technology modernization workflows, especially in industries like financial services that still run critical systems on older code. (mckinsey.com) The through line across these publications is consistent: fewer pilots, more redesign. QuantumBlack’s playbook is less a celebration of autonomous agents than a checklist for the data, controls, and operating model companies need before those agents can be trusted in production. (mckinsey.com)
