Copy article

McKinsey’s 20,000 "AI Employees": The Era of "Bot Shopping" Has Begun

ended 18. January 2026

McKinsey’s CEO Bob Sternfels has announced the firm now "employs" 20,000 AI agents alongside its 40,000 humans. The vision? A 1:1 ratio of bots to bodies within 18 months.

While the tech press applauds the innovation, I am looking at the business model. I believe we are watching a cynical and fundamental pivot in how consultancies operate. We are moving from "body shopping", the old model of billing clients for armies of junior staff, to "bot shopping," a new model where they bill for armies of proprietary code.

In the body shopping era, consultancies thrived by embedding people into your culture. It was lucrative, but it had a flaw: people leave. Junior consultants eventually quit, go in-house, or change careers.

In this new bot shopping era, they'll find a way to embed proprietary agents into your technology stack. Unlike people, proprietary code doesn't quit. It stays, it scales, and it creates a dependency that is nearly impossible to unpick.

How long before we see a new line item on the monthly project invoice?

  • Senior AI Associate: £2,500 per day
  • Availability: 24/7
  • Notes: Does not sleep, does not complain, hallucinates occasionally.

It sounds like a joke, but the economic incentive is serious. If a consultancy can charge "outcome-based fees" or "technology access retainers" for these 20,000 agents, they have successfully industrialised their revenue stream.

This raises a serious concern for industries like Finance, Insurance, and HR who rely on accurate decision-making. We are looking at an industrialised deployment of a technology that fails more often than it succeeds.

According to a 2025 MIT report, 95% of GenAI pilots fail to deliver measurable financial returns. Even BCG's own research admits that 74% of companies struggle to get AI to create real value.

Here is the danger for the client:

If a junior consultant makes a mistake, the firm apologises and retrains them.

If a proprietary "AI Employee" hallucinates—denying a mortgage application or messing up a payroll run, who pays?

My prediction is that it won't be a warranty fix. It will be a new, billable project to "investigate complex data integration issues." They won't blame their agent; they will blame your "legacy data." It is the perfect business model: the consultancy gets paid to install the bot, and then paid again to fix the chaos it creates.

We'd like your views:

  • If a consultancy's AI agent helps process applications but hallucinates a rejection, does the liability sit with them or you?
  • If "digital employees" are doing the work of staff, should their costs be capped? Or are we about to see "AI inflation" on service fees?
  • Are you willing to pay senior consultant rates for output generated by a bot?
  • Are you concerned that "hybrid workforces" are actually a Trojan horse for permanent vendor lock-in?

6 responses from the Newspage community

Copy all

Star Quote
Copy

Rohit Parmar-Mistry
Founder at Pattrn Data
Sternfels wants a 1:1 ratio of bots to bodies. This isn't innovation; it’s the industrialisation of revenue extraction. We are witnessing the birth of the "Replicant Consultant", agents that mimic professional output but lack the human judgment you actually pay for.

Let’s be clear: they aren't replacing the billable hour; they are stacking a new tax on top of it. You will now pay for the human to manage the project, plus a premium "technology access fee" for the agent to generate the outputs. It’s the perfect double-dip.

Even worse, this is a Trojan Horse for IP leakage. If these proprietary agents operate on your internal data, they may learn from you. You are effectively paying a consultancy to train their models on your trade secrets, which they will then repackage and sell to your competitors as "benchmarked insights."

It’s corporate Westworld, shiny, impressive, but a trap. Why pay premium strategic rates for a commodity bot you could license yourself for pennies?
Copy

Editor
Kate Underwood
Founder & Chief People Strategist at Kate Underwood HR and Training
Trying to leave one of these “hybrid workforce” setups will make getting out of a mobile or broadband contract look like child’s play. At least with telecoms you know what you’re trapped by: an end date, an early exit fee, a router you can post back. With consultancy-led AI, the lock-in is invisible. Their agents get threaded through your workflows, data, templates, dashboards, decision rules. It becomes “how we do things”, not just a supplier. And when you finally say “we’re switching”, they won’t call it lock-in. They’ll call it “complex integration risk” and “knowledge transfer”. Translation: another bill. Probably a big one.
It’s not the tech that traps you; it’s the dependency. The minute your process relies on their proprietary tools, you’re not buying support anymore. You’re renting part of your operating system.
Copy

Editor
Mitali Deypurkaystha
AI Governance Strategist & CMO at AISensa
The idea that ‘it’s all about data’ isn’t wrong, but what 2025 has exposed is how unprepared most organisations actually are. I’m repeatedly brought in to ‘implement AI’, only to spend weeks or months first untangling fragmented, outdated or ungoverned data. AI built on messy data doesn’t become intelligent. It becomes confidently wrong.

My concern with large consultancies racing to sell ‘AI employees’ is that they’ll treat bots as a panacea for all business ills, not a precision tool. When those agents inevitably hallucinate or fail, the blame won’t sit with the technology they sold. It will be deflected onto the client’s ‘legacy data’. That creates a perfect profit-making loop: sell the bot, then sell the clean-up.

It’s like installing a Formula One engine into a car with a cracked chassis, then charging extra when it crashes. Responsible AI means fixing the foundations first, and if you failed to do that, taking accountability when things go wrong.
Copy

Kundan Bhaduri
Entrepreneur, Investor and Landlord at The Kushman Group
The old model of renting out junior staff had a flaw because people eventually leave but proprietary code embeds itself into your infrastructure and gets what is called as a vendor lock-in.

This pivot is obviously lucrative for the consulting firms but perilous for clients given that recent data suggests 95% of generative AI pilots fail to deliver financial value while even BCG admits three quarters of companies struggle to see a return.
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
McKinsey's promising 24/7 AI agents that never sleep or complain. What they're not promising: agents that know when to shut up, escalate to humans, or admit they're guessing.

The "set and forget" fantasy ignores a fundamental flaw: these systems don't know what they don't know. A junior consultant who's uncertain asks for help. An AI agent confidently hallucinates and keeps billing. If your agent processes 10,000 decisions overnight, how many hallucinated edge cases are embedded before anyone notices? What is the opportunity cost of unpicking these sort of mistakes?

When that agent makes a £50,000 payroll mistake, does McKinsey's contract include liability? or just a slippery clause about "known AI limitations"? And who explains to staff why they're missing money they never received, or spent months ago?

This isn't innovation. It's industrial-scale overconfidence with no off switch and a monthly fee.

The business world needs a long debate about the tech emperor's new clothes.
Copy

Patricia McGirr
Founder at Repossession Rescue Network
Clients think they are paying for judgement, assurance and accountability. In reality, they are buying access to proprietary systems where responsibility is contractually blurred and risk is quietly pushed onto the customer. These AI agents are priced like senior consultants but governed like tools. When something goes wrong, it is reframed as a data or integration issue, not a professional failure, and the consultancy still invoices. Trust breaks when fees are detached from responsibility. If firms want to monetise AI as expertise, they must stand behind outcomes like experts. Otherwise this is not innovation. It is vendor lock-in with a very expensive shrug.