Bob Sternfels, the guy running McKinsey, dropped the numbers at CES: 40,000 humans, 25,000 AI agents. Working together. Right now. He wants parity by year-end.
Eighteen months ago, they had a few thousand agents. The math is aggressive.
The productivity claim
Here's what they're saying internally: 1.5 million hours saved last year on search and synthesis alone. The agents churned out 2.5 million charts in six months. That's the grunt work that used to keep junior consultants busy until 2 AM.
Sternfels laid this out during a live taping of the All-In podcast at CES with Jason Calacanis and Hemant Taneja from General Catalyst. The framing was optimistic, as you'd expect. But buried in the details: back-office roles are shrinking by 25%. Client-facing roles are growing by the same amount.
The back-office output? Up 10%, despite having fewer people.
Make of that what you will.
What these agents actually do
They're not chatbots. According to the HBR IdeaCast interview, these are autonomous systems handling entire job functions. Research synthesis. Document compilation. Data analysis. The whole workflow, not just individual tasks.
QuantumBlack, McKinsey's 1,700-person AI division, drives all of this. That team now accounts for 40% of the firm's work. Two years ago, AI was a footnote. Now it's nearly half the business.
The internal platform is called Lilli. More than 70% of employees use it monthly, running half a million prompts. Each query apparently eliminates about six minutes of manual document hunting. That adds up.
The hiring pivot nobody's talking about
Sternfels made an interesting argument about credentials. Forget where candidates went to school, he said. Look at their GitHub portfolio. Look at demonstrated skills.
"Let's actually get to the content," was his line. Could different pathways open up for people who would have been overlooked before?
Maybe. But McKinsey isn't exactly known for democratizing access to elite jobs. The firm has always recruited from the same handful of schools. Whether AI actually changes that culture remains to be seen.
What they're definitely looking for: people who can be "great McKinsey consultants and great technologists" simultaneously. Both. Not one or the other.
The layoff question
Here's the context that didn't make the CES presentation. McKinsey's headcount peaked around 45,000 in 2022. It's now closer to 40,000. Bloomberg reported in December that additional cuts of up to 10% in non-client-facing roles are being discussed for the next 18-24 months.
The firm's official line: they're "aligning capabilities with evolving priorities." Translation varies depending on who you ask.
Revenue has hovered around $15-16 billion for a couple years. Flat. Clients are more cautious about big consulting bills. And some of the work that would have gone to McKinsey is now being done by companies' own internal teams using the same AI tools.
What BCG and Bain are doing
McKinsey isn't alone in this. BCG went from zero to 20% of revenue from AI services in about two years, roughly $2.7 billion. They've hired 1,000 new staff specifically for AI work. Their AgentKit framework lets domain-specific agents collaborate on complex problems.
Bain partnered with OpenAI and equipped 13,000 consultants with ChatGPT Enterprise. Their employees built over 19,000 custom GPTs. That's a lot of experimentation.
But nobody else has published numbers quite like McKinsey's 25,000 agents. Whether that's because McKinsey is ahead or because they're better at marketing the shift, I couldn't say.
The business model change
This part is actually interesting. Sternfels said McKinsey is moving away from the traditional advisory model. Instead of fee-for-service consulting, they want to identify joint business cases with clients and help underwrite the outcomes.
That's a fundamentally different relationship. It also means more risk for McKinsey if the advice doesn't work. Whether that's a genuine strategic shift or a way to compete with tech-native consultancies that have been eating their lunch on implementation work, the direction is clear.
Three skills AI can't replicate
Sternfels identified what he thinks will matter for new graduates:
Setting the right aspiration. Do you aim for low Earth orbit or Mars? That's a uniquely human call.
Judgment. AI predicts based on patterns. It doesn't know when to break them.
True creativity. The "orthogonal thinking" that disrupts established norms.
Convenient that these are exactly the skills McKinsey has always claimed to sell. Less convenient: the junior consultants who traditionally developed these skills by doing the research and synthesis work that's now automated.
How do you learn judgment if you never do the grunt work that builds context? McKinsey hasn't answered that yet.
What happens next
Sternfels expects the firm to have equal numbers of humans and AI agents by December 2026. That's 40,000 and 40,000, if the math holds.
The half-life of skills is shrinking, he noted. Return on investment for employee training used to be about seven years. Now it's 3.6 years. That number is only getting shorter.
Whether that's a warning or just context depends on your perspective.
