How McKinsey Plans to Survive AI and Reinvent Consulting (Category AI)

An interesting half-hour episode of HBR IdeaCast, with Adi Ignatius interviewing Bob Sternfels, McKinsey’s Global Managing Partner and effectively its CEO. The central idea is that AI is 30–50% technology; the rest is organizational design and processes. Attaching an LLM to an old process yields local automation rather than a system-wide increase in throughput.

I watched with interest and took away several ideas that also apply to software development.

1️⃣ The barrier is organizational, not just the model Silos, unnecessary handoffs, and approval layers eat into AI’s benefits more than prompt quality does.

2️⃣ Human + agent is the new unit of productivity McKinsey talks about tens of thousands of agents and moving toward 1 agent per person. The interview does not define “agent,” so it is unclear what to compare those figures with. Agents are being counted as part of the workforce, with their growing numbers highlighted.

3️⃣ As analytics becomes a commodity, value moves to outcomes When analysis becomes cheaper, the advantage goes to those who deliver results and accept responsibility for the effect. The same applies inside companies: focus less on tickets closed and more on outcomes improved.

4️⃣ Hiring and growth: resilience and learning matter more than perfect grades McKinsey’s hiring criteria are changing. It once sought top students, but learned that overcoming crises predicts success better than excellent grades. Prefer people who can fall, get back up, learn, and work with others.

5️⃣ Speed wins even with more mistakes, provided recovery is good This directly parallels sound engineering practices: small batches, feature flags, observability, and fast rollbacks reduce the cost of mistakes and allow faster work.

Advice for engineers

1. Stop treating AI as a magic button. Build a pipeline: agent → PR → quality gates → human review. Tests, linters, and security scans still apply.
2. Find where an agent removes a handoff rather than merely writes text faster. Start with the end-to-end value stream.
3. Develop second-wave skills: judgment, alternatives, and counterexamples. Ask the model for risks, checks, and what could go wrong, not just an answer.

Advice for technical leaders

1. Treat AI as operational transformation, rather than a collection of pilots.
2. Choose 2–3 processes and redesign them end to end, for example triage → fix → deploy.
3. Measure outcomes and balance speed with stability: DORA’s lead time, deployment frequency, MTTR, and change-failure rate, plus DevEx/SPACE to prevent acceleration from burning out the team.
4. Manage agents as products, with inventory, ownership, and auditing: who owns each agent, what data it can see, what tools it calls, and which gates it must pass.

#AI #Engineering #Leadership #Devops #Productivity #Management