McKinsey & Co Now Employs 25,000 AI Agents—And That’s Just the Start
McKinsey's deployment of 25,000 AI agents in 18 months signals the end of experimental pilots and the beginning of agent-first enterprise operations
When McKinsey CEO Bob Sternfels casually mentioned this month that his firm now has 25,000 AI agents on the payroll, he wasn’t making a prediction. He was describing what’s already happening.
Eighteen months ago, McKinsey used a few thousand agents. Now they’re at 25,000, working alongside 40,000 humans. Do the maths—that’s already a workforce that’s 40% non-human. And Sternfels wants every single human employee paired with at least one agent within the next year and a half.
The numbers are staggering. Those agents generated 2.5 million charts in six months. The firm saved 1.5 million hours last year just on search and synthesis tasks. This isn’t some pilot programme tucked away in a corner of the business. McKinsey’s QuantumBlack division—1,700 people who drive all its AI work—now accounts for 40% of the firm’s work.
Think about that. The world’s most prestigious consulting firm has restructured nearly half its business around AI in less than two years.
It turns out McKinsey’s experience lines up almost exactly with predictions from Nexos.ai, the Lithuanian company founded by the people behind Nord Security. Their 2026 forecast indicates we’re past the chatbot experimentation phase. What’s coming is specialised agent fleets that function as actual team members, not toys that employees play with once and forget.
Žilvinas Girėnas, Nexos’s head of product, puts it bluntly: “Businesses are no longer deploying one agent to solve one problem. They’re building teams of specialised agents that work together.”
Payhawk, the fintech company, deployed Nexos’s platform across finance, support, and operations. They cut security investigation time by 80%, hit 98% data accuracy, and reduced processing costs by 75%. Zero compliance violations. These aren’t marginal gains—they’re the kind of numbers that get CFOs interested.
From Generic Chatbots to Named Colleagues
Here’s what changed. Companies spent 2024 rolling out ChatGPT Enterprise or Copilot, only to see adoption fizzle after the initial novelty wore off. Generic tools get generic results.
Now they’re creating named agents with specific jobs. HR has an agent that screens CVs against the team’s actual criteria. Legal has one that reviews contracts according to standards they’ve defined. Sales has agents drafting emails in the cadence that works for their market.
The difference isn’t subtle. Move from one general chatbot to multiple purpose-built agents, and adoption jumps dramatically. People actually use tools that solve their specific problems.
McKinsey’s deployment proves this. Their agents aren’t doing vague “AI stuff”—they’re handling routine analytics, research synthesis, and data compilation. The humans focus on client relationships and strategic thinking, which is what McKinsey actually gets paid for.
The firm’s also fundamentally changing how it charges. They’re moving away from traditional advisory fees toward outcome-based compensation tied to joint business cases. That only works if you’ve got the infrastructure to deliver at scale. Which means agents.
The Platform Consolidation Problem
Nexos predicts teams will consolidate on a single platform rather than juggling multiple agent tools. This is already happening, and it’s happening for boring operational reasons.
Deploy five to ten agents across different platforms, and you get spiralling costs, fragmented security, and the administrative nightmare of managing separate logins and billing. Research shows unified platforms deliver 2x faster deployment and much better visibility into what’s actually being used.
Girėnas makes the practical case: “When teams juggle 10 agents across different tools, they stop using half of them. With a single platform, they can fully leverage all their agents.”
It’s not sexy, but it’s how enterprise software actually gets adopted. Nobody wants to manage ten different vendor relationships for what should be integrated functionality.
Who Actually Owns This Stuff?
The ownership shift is more interesting. IT departments typically drive technology rollouts, but AI agents are increasingly being configured and managed by business teams.
HR heads set up their screening agents. Legal teams version their contract review playbooks. Sales leaders optimise their outreach templates. They share what works across departments without waiting for engineering to build everything from scratch.
This makes sense. These people know their workflows. They understand what good output looks like. Platforms that let them configure agents without writing code are the ones seeing traction.
McKinsey’s hiring accordingly. Sternfels wants people who can be “great McKinsey consultants and great technologists.” Boston Consulting Group created “forward-deployed consultants” who build AI tools directly on client projects. The model is changing—you can’t just be good at strategy or good at tech anymore.
The Supply Problem Nobody’s Talking About
Here’s where it gets messy. Teams deploy two or three successful agents, and suddenly every department wants the same capabilities. Marketing wants workflow automation. Finance needs compliance agents. Customer success wants support tools.
Building custom agents takes time. Industry estimates suggest 40% of enterprise software will have embedded AI agents by the end of 2026, up from under 5% in 2024. That’s a huge gap, and there aren’t enough engineers to fill it.
“The answer is template libraries and prebuilt agents that teams can adapt quickly.” “The teams that succeed will be equipped with agent libraries rather than creating every agent from scratch, Demand is rising fast, and the only way to keep pace is through templates.”
Žilvinas Girėnas, Nexos’s head of product
McKinsey’s timeline supports this. Scaling from a few thousand agents to 25,000 in eighteen months requires systematic deployment approaches, not bespoke development for every use case.
What This Means for Marketing and Media
The implications for advertising and publishing are direct. Campaign management, audience targeting, creative testing, and performance optimisation—all repetitive processes that agents can handle.
Media organisations experimenting with AI content tools are moving from pilots to production. Some of this works, some doesn’t. But the direction is clear.
McKinsey’s research estimates AI agents could unlock $2.9 trillion in annual US economic value by 2030. That only happens if companies redesign workflows around what humans and agents each do best, not just automate isolated tasks.
Early examples show the pattern. Sales teams let agents prioritise leads and manage outreach, freeing them for negotiation and relationship-building. Customer service agents handle routine inquiries while people tackle complex or sensitive cases.
The firms getting this right aren’t treating it as an IT project. They’re treating it as a business transformation. BCG, PwC, and the other big consultancies are shifting from traditional advisory work to multi-year AI transformation projects. The tools they sell to clients are the same ones they use internally.
The Reality Check
Training people to work with agents isn’t trivial. Integration across legacy systems creates technical debt. Governance frameworks are still being figured out. Security teams are managing risks they’re only beginning to understand.
And yet the momentum is unmistakable. Companies watching competitors deploy agents successfully face pressure to accelerate their own timelines. The technology keeps advancing—things that seemed experimental six months ago are now reliable enough for production use.
McKinsey’s aggressive deployment schedule sets a benchmark. From a few thousand agents to 25,000 in eighteen months, targeting one-plus agents per employee within another year and a half. Other enterprises will struggle to ignore that pace.
Nexos’s predictions for 2026—specialised agent teams, platform consolidation, business-led deployment, supply constraints—aren’t particularly bold given McKinsey’s current reality. They’re describing what’s already underway.
The question for most organisations isn’t whether this is coming. It’s whether they can deploy effective agent fleets quickly enough, integrate them properly, and build governance that scales. Based on what’s happening at McKinsey, Payhawk, and the consulting giants, answers are arriving faster than most executives expected.
The transition from experimental chatbots to operational colleagues is happening. Companies figuring this out early position themselves for an advantage. Those still treating AI as a side project may find they’re already behind.
